Double-material precise spreading dynamic adjusting method and system based on advancing speed

Abstract

The invention relates to the technical field of precision agriculture, and discloses a double-material precise broadcast application dynamic adjusting method and system based on the advancing speed, and the method comprises the steps: collecting original data, and carrying out the preprocessing; initial material characteristic parameters are tested, double-material target flow is calculated, online identification is carried out, and the material characteristic parameters are continuously updated; constructing a mechanism inverse model to calculate a feedforward control instruction, calculating a feedback control quantity to obtain a comprehensive control instruction, and superposing predictive compensation and dead zone compensation; establishing a coupling model to eliminate inter-channel interference, calculating a ratio deviation and performing multi-objective optimization, and scheduling priorities according to the saturation state of each channel; recording data of the whole operation process, calculating quality indexes, establishing a control strategy model for control decision optimization, and establishing a multi-working-condition parameter library; according to the invention, the seeds and the fertilizer can be uniformly distributed in the field according to the target dosage and proportion under the operation condition that the advancing speed continuously fluctuates.

Description

Technical Field

[0001] This invention relates to the field of precision agriculture technology, and more specifically, to a dynamic adjustment method and system for precise application of dual materials based on travel speed. Background Technology

[0002] As modern agriculture develops towards precision and intelligence, precision fertilization and precision sowing technologies have become key means to increase crop yields, reduce production costs, and minimize environmental pollution. During mechanized agricultural operations, the speed of the machinery changes in real time due to various factors such as terrain variations, soil conditions, and driver operation. Traditional fixed-rate application methods cannot adapt to these dynamic changes, resulting in uneven distribution of materials in the field, which wastes resources and affects crop growth quality. Especially in scenarios requiring the simultaneous application of both seeds and fertilizers, accurately adjusting the application rate of both materials based on real-time changes in speed to ensure even distribution according to predetermined ratios and target dosages is a key technical challenge that needs to be addressed in the field of precision agriculture.

[0003] Traditional agricultural machinery spreading devices primarily employ mechanical transmission or simple electronic control, operating by pre-setting fixed spreading parameters. This approach has significant limitations. When the machinery's speed increases, the area covered per unit time increases. If the spreading rate remains constant, the amount of material spread per unit area decreases accordingly; conversely, when the speed decreases, the amount of material spread per unit area increases. This mismatch leads to severely uneven material distribution in the field, resulting in inconsistent fertilization or sowing density, directly impacting crop germination, growth, and final yield. In dual-material spreading scenarios, the two materials often have different physical properties, such as particle size, density, and flowability, making spreading control even more complex. If the spreading rates of the two materials cannot be precisely coordinated, the problem of uneven distribution will be further exacerbated.

[0004] While some solutions for variable fertilization and variable seeding have emerged in existing technologies, most primarily focus on spatial variable control based on differences in soil nutrients or crop growth status—that is, adjusting the application rate according to the needs of different locations in the field. They pay insufficient attention to dynamic changes over time during operations, particularly the instantaneous changes in application demand caused by fluctuations in travel speed. Although some technologies have introduced speed detection functions, their control strategies are relatively simple, with slow response times, making it difficult to achieve real-time tracking and precise compensation for rapid changes in travel speed. Furthermore, most existing technologies are designed for single materials, lacking a systematic solution for the simultaneous and precise application of two materials. The application control of the two materials is independent, lacking a coordination mechanism, and failing to guarantee that the ratio between the two materials remains stable during dynamic operations. Summary of the Invention

[0005] This invention provides a dynamic adjustment method and system for precise application of two materials based on travel speed, solving the technical problem in related technologies that cannot maintain the accuracy and ratio stability of application of two materials when speed fluctuates.

[0006] This invention provides a dynamic adjustment method for precise application of two materials based on travel speed, comprising: Raw data from three speed sensors and a dual-channel material flow sensor are collected and preprocessed to obtain an operational status perception dataset. Based on the operational status awareness dataset, test the initial parameters of material characteristics, calculate the target flow rate of dual materials, perform online identification and continuous update of material characteristic parameters, and obtain the material characteristic adaptive modeling result set; Based on the adaptive modeling result set of material properties, the mechanism inverse model is constructed to calculate the feedforward control command, the feedback control quantity is calculated to obtain the comprehensive control command, and the predictive compensation and dead zone compensation are superimposed to obtain the comprehensive control command set of the dual-channel actuator. Based on the integrated control instruction set of the dual-channel actuator, a coupling model is established to eliminate interference between channels, the ratio deviation is calculated and multi-objective optimization is performed, and the coordinated and optimized control instruction set is obtained according to the scheduling priority of each channel's saturation state. Based on the coordinated and optimized control instruction set, the data of the entire operation process is recorded and the quality indicators are calculated. A control strategy model is established to optimize control decisions, a multi-condition parameter library is established, and a set of continuous improvement and optimization results is obtained.

[0007] In a preferred embodiment, the step of obtaining the job status awareness dataset includes: The first speed signal is obtained based on the wheel speed sensor. When the difference between the wheel acceleration and the overall machine acceleration exceeds the preset slippage threshold, the confidence weight of the wheel speed sensor is reduced. The speed obtained from the second speed signal based on the GPS positioning module is verified by numerically differentiating the position coordinates and comparing it with the speed directly output by the GPS. The third velocity information is obtained by integrating the acceleration signal using an inertial measurement unit (IMU). The federated Kalman filter algorithm is used to fuse the three velocity information. The bottom layer consists of three local filters that process the three velocities respectively, and the top layer main filter integrates the results of the local filters to obtain the fused velocity estimate and velocity confidence.

[0008] In a preferred embodiment, the federated Kalman filter algorithm includes: For the local filter of GPS velocity, the observation noise covariance is dynamically adjusted according to the accuracy factor value output by the GPS module; The main filter performs weighted fusion based on the velocity estimates and error covariance of the local filter outputs, with the weighting coefficients calculated based on the information matrix of each local estimate.

[0009] In a preferred embodiment, the initial parameters of the tested material characteristics, the calculation of the target flow rates of the two materials, and the online identification and continuous updating of the material characteristic parameters specifically include: Before the operation begins, perform material characteristic tests and control the seed and fertilizer metering device to perform step response and frequency response tests to identify material flow response characteristics. The least squares method is used to identify the estimated values ​​of the characteristic parameters of the actual material based on the test data; Based on the travel speed, operating width, target unit area application rate and material characteristic parameters, the speed-flow mapping model is used to dynamically calculate the target instantaneous application flow rate. Introduce feedforward compensation for velocity change rate; when a velocity change trend is detected, add feedforward compensation amount to the target flow rate. The recursive least squares method is used for online updating of material property parameters, and a forgetting factor mechanism is introduced.

[0010] In a preferred embodiment, the constructing mechanism calculates the feedforward control command using an inverse model, calculates the feedback control quantity to obtain the comprehensive control command, and superimposes predictive compensation and dead-zone compensation, including: A dynamic transmission relationship model of the actuator from control commands to material output flow rate is established, and a nonlinear relationship model between seed and fertilizer discharge flow rate and rotation speed is established as a positive dynamic model. Based on the forward dynamic model, an inverse model is constructed to solve the inverse function of the static nonlinear relationship between flow rate and rotational speed, and a first-order advance compensation element is constructed for the first-order inertial element. The target flow rate is input into the inverse model to obtain the feedforward control command; additional feedforward compensation is introduced by the prediction of speed change, and the expected flow rate change is multiplied by the prediction compensation coefficient and then superimposed on the feedforward control command. Based on the deviation between the actual flow rate and the target flow rate, a nonlinear PID feedback control algorithm is used to calculate the feedback control quantity, and a nonlinear proportional gain is introduced to adaptively adjust according to the magnitude of the deviation. Integral separation and integral limiting mechanisms are introduced. Integration is paused when the absolute value of the deviation exceeds the preset integral separation threshold, and integration is stopped when the control quantity reaches saturation. Dead zone compensation and friction compensation torque are superimposed on the control commands.

[0011] In a preferred embodiment, constructing the inverse model based on the forward dynamic model includes: A quadratic polynomial is used to describe the relationship between flow rate and rotational speed. The coefficients are obtained by fitting the actual flow rate measured at different rotational speeds through bench calibration tests. Solving the inverse function of the nonlinear relationship between flow rate and rotational speed yields the mapping relationship between rotational speed and flow rate; The dynamic response of the servo motor drive system is described by a first-order inertial element, and a first-order lead compensation element is constructed as the dynamic inverse model. The static inverse model and the dynamic inverse model are connected in series to form a complete inverse model.

[0012] In a preferred embodiment, the step of establishing a coupling model to eliminate inter-channel interference, calculating the allocation deviation and performing multi-objective optimization, and scheduling priority according to the saturation state of each channel includes: A coupling characteristic model is established, and an experimental identification method is used to fix the control quantity of one channel and apply a step change control quantity to another channel. The coupling transfer function is then obtained by fitting. Calculate the coupling degree index, and activate the decoupling control mode when the coupling degree exceeds the preset coupling degree threshold; A feedforward decoupling control strategy is adopted to calculate the decoupling compensation amount. The control command of the other channel is input into the coupling transfer function to obtain the coupling interference amount. The compensation amount is the inverse of the interference amount. The allocation deviation is calculated based on the actual flow of the two channels. When the allocation deviation exceeds the preset allocation deviation start threshold, the allocation coordination control mode is activated. A multi-objective optimization function is established, and the objectives are summed with weighted coefficients. The optimal control quantity is then solved using a constrained optimization algorithm. Channel priority is defined and dynamically calculated based on flow deviation, deviation duration, and control saturation level. When saturation is detected in one channel, the ratio is corrected by adjusting the control quantity of another channel.

[0013] In a preferred embodiment, establishing the multi-objective optimization function includes: The objectives of the multi-objective optimization function include minimizing two flow deviations: minimizing flow ratio deviation and optimizing the smoothness of control command changes; the weight coefficients of each objective are dynamically adjusted according to the magnitude of the ratio deviation. Constraints include upper and lower limits of control variables and control variable rate of change constraints. When one channel is saturated, the target flow rate of the other channel is calculated by back-calculating the actual flow rate of the saturated channel and the target ratio.

[0014] In a preferred embodiment, the steps of recording data throughout the entire operation process and calculating quality indicators, establishing a control strategy model for control decision optimization, and establishing a multi-condition parameter library include: The data collected during the operation is stored in a time-series format to form an operation data log; After the operation is completed, calculate the average deviation, maximum deviation, and standard deviation of the flow rate, and calculate the average, maximum, and standard deviation of the proportion deviation. The dynamic characteristics of the control system are evaluated using frequency domain analysis based on operational data, and the bandwidth frequency and resonant frequency of the system are identified. A data-driven control strategy model is established using machine learning methods. A neural network model is built, with the network input including the current speed, the rate of change of speed, the target flow rate, the actual flow rate, and the historical sequence of flow deviation. The network output is the optimal control variable. The control parameters are optimized using an automatic parameter tuning algorithm, and a multi-condition parameter library is established for different operating conditions.

[0015] This invention provides a dynamic adjustment system for precise application of dual materials based on travel speed, used to execute the aforementioned dynamic adjustment method for precise application of dual materials based on travel speed, comprising: The operation status sensing module is used to collect raw data from three speed sensors and two material flow sensors and preprocess it to obtain the operation status sensing dataset. The material property adaptive modeling module, based on the operation status perception dataset, tests the initial parameters of material properties, calculates the target flow of dual materials, performs online identification and continuous updates of material property parameters, and obtains the material property adaptive modeling result set. The dual-channel actuator control module constructs an inverse model of the mechanism based on the adaptive modeling result set of material characteristics, calculates feedforward control instructions, calculates feedback control quantities to obtain comprehensive control instructions, and superimposes predictive compensation and dead zone compensation to obtain a comprehensive control instruction set for the dual-channel actuator. The coordinated optimization control module, based on the integrated control instruction set of the dual-channel actuator, establishes a coupling model to eliminate interference between channels, calculates the ratio deviation and performs multi-objective optimization, and obtains the coordinated optimized control instruction set according to the scheduling priority of each channel's saturation state. The continuous improvement and optimization module records data from the entire operation process and calculates quality indicators based on the coordinated and optimized control instruction set. It establishes a control strategy model to optimize control decisions, builds a multi-condition parameter library, and obtains a continuous improvement and optimization result set.

[0016] The beneficial effects of this invention are as follows: By acquiring high-precision travel speed information in real time through multi-sensor fusion technology, dynamically calculating the target application rate based on an adaptive model of material characteristics, and driving the dual-channel application actuator with a nonlinear compensation and feedforward-feedback composite control strategy, the system achieves precise synchronous application of two materials using decoupled control and coordination optimization techniques. This effectively solves the technical problem of uneven material application caused by fluctuations in the travel speed of the application machinery. Even under operating conditions with continuously fluctuating travel speed, the system ensures that seeds and fertilizers are evenly distributed in the field according to the target dosage and ratio, significantly improving the quality of sowing and fertilization, and enhancing crop emergence uniformity and plant quality.

[0017] By employing a federated Kalman filter algorithm to fuse multiple speed information streams, the accuracy and reliability of speed measurement are improved. Online parameter identification and self-learning mechanisms enable the control model to adapt to different material characteristics and operating conditions. The introduction of feedforward compensation and predictive control strategies effectively reduces control lag and improves the system's responsiveness to rapid speed changes. Decoupled control and coordinated optimization achieve precise maintenance of the dual-material application ratio. A data-driven intelligent learning mechanism enables continuous optimization of control performance. This invention can effectively improve agricultural production efficiency, reduce material input costs, and minimize environmental pollution caused by excessive application, thus playing a significant role in promoting sustainable agricultural development. Attached Figure Description

[0018] Figure 1 This is a flowchart of the main process of a dynamic adjustment method for precise application of dual materials based on travel speed according to the present invention. Figure 2 This is a detailed flowchart of a dynamic adjustment method for precise application of dual materials based on travel speed according to the present invention; Figure 3 This is a block diagram of a dynamic adjustment system for precise application of dual materials based on travel speed, according to the present invention. Detailed Implementation

[0019] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0020] At least one embodiment of the present invention discloses a dynamic adjustment method for precise application of dual materials based on travel speed, such as... Figures 1 to 2 As shown, it includes: Step 1: Collect raw data from three speed sensors and a dual-channel material flow sensor and preprocess it to obtain the operation status perception dataset. Specifically, the following steps are included: Step 1.1: Acquisition and preprocessing of multi-channel speed sensor signals; Based on a photoelectric encoder speed sensor mounted on the drive wheel axle of the work implement, a pulse counting method is used to measure the wheel rotation speed in real time, obtaining the first speed signal. The photoelectric encoder outputs 3600 pulses per revolution. The controller sets a timer to read the pulse count value at a frequency of 100Hz. By calculating the difference in the number of pulses between two adjacent readings, combined with the wheel circumference and encoder resolution, the travel distance is calculated, and then the instantaneous travel speed is calculated based on the time interval. Specifically, the pulse count difference is first converted into the number of wheel revolutions, then multiplied by the wheel circumference to obtain the travel distance, and finally the travel distance is divided by the time interval to obtain the instantaneous speed. To address the possibility of wheel slippage, a slippage detection mechanism is introduced into the calculation process. By monitoring the consistency between the wheel acceleration and the overall machine acceleration, when the difference between the two exceeds 0.5 m / s², slippage is determined. At this time, the confidence weight of the wheel speed sensor is reduced from the normal value of 0.4 to 0.1. After the slippage is eliminated, the weight is gradually restored to avoid abrupt changes in speed estimation caused by sudden weight changes.

[0021] Based on a high-precision differential GPS positioning module, RTK real-time dynamic differential positioning technology is used to obtain centimeter-level accurate position coordinates and speed information directly output by GPS, serving as the second speed signal. The GPS module updates position and speed data at a frequency of 10Hz. After receiving the GPS data, the controller performs coordinate system transformation, converting latitude and longitude coordinates into a local Cartesian coordinate system with the starting point of the work area as the origin. Numerical differentiation is performed on continuous position coordinates to calculate the distance between adjacent position points. This distance is then divided by the time interval. To reduce the noise impact of numerical differentiation, a three-point smoothing differentiation method is used to calculate the speed. This method uses adjacent position points before and after the current moment, and divides the difference between the position at the next moment and the position at the previous moment by twice the sampling period to obtain the speed at the current moment. This method is equivalent to first performing three-point smoothing filtering on the position data and then calculating the differentiation to obtain the speed based on the position differentiation. This speed is then cross-validated with the speed directly output by GPS. When the difference between the two is small, it indicates good GPS signal quality. When the difference is large, it indicates possible multipath interference or signal obstruction, in which case the reliability weight of the GPS speed is reduced.

[0022] Based on an Inertial Measurement Unit (IMU), a three-axis accelerometer and a three-axis gyroscope are used to measure the acceleration and angular velocity of the working machine in real time, obtaining a third channel of speed information. The IMU outputs data at a high frequency of 200Hz, and the controller integrates the acceleration signal to obtain the speed information. Due to the cumulative error of acceleration integration, simple IMU speed estimation will produce drift. Therefore, IMU speed is mainly used for high-precision speed change detection over short periods, especially during rapid speed changes, where the IMU can provide a faster response than GPS and wheel sensors. To address the IMU's zero-point drift problem, the controller performs static calibration before operation, recording the acceleration and angular velocity outputs in a stationary state as the zero-point reference. Considering that the IMU zero point will slowly drift with temperature and time, a dynamic zero-point estimation method is used during operation. When the entire machine is detected to be stationary, for example, if the speed is below 0.1 m / s for 5 consecutive seconds, it is determined to be stationary. At this time, the IMU's output acceleration should theoretically be zero, and the actual output value is the current zero-point deviation, which is used to update the zero-point reference. The determination of a stationary state integrates multiple information such as wheel speed, GPS speed, and engine speed to avoid misjudgment. The dynamically updated zero-point reference adopts an exponential smoothing filter method. The new zero point is obtained by weighted averaging of the old zero point and the current measurement value, where the old zero point has a weight of 0.9 and the current measurement value has a weight of 0.1, achieving a smooth transition.

[0023] The raw data from the three speed sensors undergoes time synchronization processing. Since the sampling frequencies and data output times of the three sensors differ, a unified time reference needs to be established. The controller uses a hardware clock as the global time reference, assigning a precise timestamp to each sensor's data. During data fusion, interpolation alignment is performed based on the timestamps to ensure that the fused values ​​are from the same moment. Hardware anti-jitter processing is applied to the pulse signals from the wheel speed sensors to eliminate spurious pulses caused by mechanical vibration. GPS data undergoes integrity verification, discarding invalid data with insufficient satellite counts or excessively large accuracy factors. Temperature compensation is applied to the IMU data, with real-time correction of the zero point and sensitivity based on the IMU chip temperature measured by the temperature sensor.

[0024] Step 1.2: Fusion of multi-source velocity information based on federated Kalman filtering; Based on the preprocessed data from three speed sensors, a federated Kalman filter algorithm is used for information fusion to obtain a high-precision speed estimate and speed confidence level. The federated Kalman filter adopts a hierarchical structure, with three local filters at the bottom layer to process wheel speed, GPS speed, and IMU speed respectively, and a master filter at the top layer. The results of the three local filters are combined to obtain the globally optimal estimate.

[0025] For a wheel speed sensor, a local Kalman filter 1 is established. The state variables are the wheel linear velocity and wheel acceleration, and the observed quantity is the instantaneous velocity measured by the encoder. The state transition equation adopts a uniformly accelerated motion model, i.e., the velocity at the next moment equals the current velocity plus the current acceleration multiplied by the sampling period, and the acceleration at the next moment equals the current acceleration. The observation equation describes the relationship between the encoder measurement and the actual velocity. Process noise reflects the randomness of wheel speed changes, and observation noise reflects the encoder measurement error and the quantization error of pulse counting. The filter predicts the velocity at the next moment based on the state transition equation, calculates the prediction error based on the observation equation and the actual measured value, and uses Kalman gain to correct the predicted value to obtain the optimal estimate. The magnitude of the Kalman gain reflects the degree of confidence in the prediction and measurement. When the process noise is large and the observation noise is small, the gain is large, indicating greater confidence in the measured value; conversely, the gain is small, indicating greater confidence in the predicted value.

[0026] For GPS velocity, a local Kalman filter 2 is established. The state variables are the overall velocity and acceleration, and the observations are the velocity output by GPS and the velocity based on position differential. The state transition equation describes the dynamic relationship of the overall motion, and the observation equation describes the relationship between the GPS measurement and the actual velocity. Process noise reflects the randomness of the overall velocity change, and observation noise reflects the GPS measurement error, including the combined effects of satellite geometric distribution, atmospheric delay, and multipath effects. The filter also performs a prediction-update recursive process to obtain the optimal estimate of GPS velocity. To address the time-varying characteristics of GPS signal quality, the filter dynamically adjusts the observation noise covariance based on the Position Dilution of Precision (PDOP) value output by the GPS module. When the PDOP value increases, indicating a decrease in positioning accuracy, the observation noise covariance is increased to reduce the confidence in the GPS measurement. Specifically, the noise covariance is set to 0.01 when PDOP is less than 2, 0.04 when PDOP is between 2 and 5, 0.25 when PDOP is between 5 and 10, and 1.0 when PDOP is greater than 10.

[0027] For the IMU velocity, a local Kalman filter 3 is established. The state variables are the velocity and acceleration measured by the IMU, and the observation is the accelerometer output value. The state transition equation obtains the velocity through acceleration integration, and the observation equation describes the accelerometer measurement relationship. Process noise reflects the randomness of acceleration changes and the cumulative error of integration, while observation noise reflects the measurement error of the accelerometer. The filter uses the dynamically updated zero-point reference from step 1.1 to correct the acceleration measurement value and suppress drift accumulation.

[0028] The main filter receives the outputs of three local filters, including velocity estimates and their corresponding error covariances. It then performs a weighted fusion of the three local estimates using a federated filtering algorithm. The weighting coefficients are calculated based on the information matrices of each local estimate. The information matrix equals the inverse of the error covariance, and the weighting coefficient equals the local information matrix divided by the sum of the three local information matrices. A smaller error covariance results in a larger information matrix and a larger weight. The fusion formula of the main filter is: the global velocity estimate equals the sum of the three local velocity estimates multiplied by their respective weighting coefficients, and the inverse of the global error covariance equals the sum of the inverses of the three local error covariances. This fusion method ensures that the global estimate's error covariance is minimized, achieving statistical optimization.

[0029] The main filter also calculates the confidence level of the velocity estimate, evaluating it based on the consistency of the outputs of the three local filters. High confidence indicates that the three local estimates are close to each other, while low confidence indicates significant differences. This confidence level, as a measure of the quality of the velocity information, is passed to subsequent control mechanisms. In control decision-making, the control strategy is adjusted based on the confidence level: aggressive control is used when confidence is high, and conservative control is used when confidence is low, thus improving system robustness. The confidence level is calculated using the standard deviation of the three local estimates as a consistency measure: a standard deviation less than 0.2 m / s indicates high confidence, a standard deviation between 0.2 and 0.5 m / s indicates medium confidence, and a standard deviation greater than 0.5 m / s indicates low confidence.

[0030] Step 1.3, Signal acquisition and processing from dual-channel material flow sensor; Based on a photoelectric seed sensor installed at the outlet of the seed metering device, a through-beam photoelectric switch is used to detect passing seed particles and obtain a seed flow monitoring signal. The photoelectric sensor consists of a transmitter and a receiver, installed on both sides of the seed metering pipe to form a beam detection line. When a seed particle passes through, it blocks the beam, reducing the light intensity at the receiver and outputting a low-level pulse signal. The controller captures the pulse signal via an external interrupt. Each captured falling edge indicates that a seed has passed. A 1-second sliding time window is used to count the number of seeds, updated every 0.1 seconds. The number of seeds in the window is converted into mass flow rate. The conversion method is as follows: first, the number of seeds is multiplied by the weight per thousand seeds to obtain the mass in grams. Since the weight per thousand seeds represents the mass of one thousand seeds, it needs to be divided by 1000 to convert it to the weight per seed, finally obtaining the seed mass flow rate in grams per second. To address potential missed or duplicate detections due to uneven seed size and varying passing speeds, the sensor is configured with appropriate response time and sensitivity threshold. The response time is set to half the typical seed passing time to avoid multiple pulses from a single seed, and the sensitivity threshold is set to the minimum amount of shading to ensure stable detection of a single seed. The optimal parameters are determined through calibration experiments.

[0031] Based on an impact-type mass flow sensor installed in the fertilizer discharge pipeline, the fertilizer flow rate is measured using the momentum principle to obtain the fertilizer flow monitoring signal. The impact-type flow sensor contains a swingable impact plate connected to a pressure sensor. Fertilizer particles, after exiting the discharge device, fall freely and impact the impact plate. The resulting impact force is proportional to the fertilizer flow rate, and the pressure sensor converts this impact force into a voltage signal output. The controller acquires the voltage signal via an analog-to-digital converter (ADC) at a sampling frequency of 1000Hz and converts the voltage value into mass flow rate according to a calibration relationship. The calibration relationship is established through bench testing. Under known flow rate conditions, the sensor output voltage is recorded, and a functional relationship between flow rate and voltage is fitted, typically a linear or quadratic function. To address the impact force fluctuations caused by uneven fertilizer particle descent, a moving average filter is applied to the acquired voltage signal. The window length is set to 100 sampling points, corresponding to a 0.1-second time window. The filter introduces a delay of approximately 0.05 seconds, which is compensated for in subsequent control. The filtered signal is smooth, eliminating the influence of instantaneous fluctuations and preserving the true flow rate trend.

[0032] To improve the reliability of flow measurement, two flow sensors based on different principles are configured for redundant monitoring of each material. An image recognition-based seed counting device is added to the seed channel. A high-speed camera captures images of falling seeds, and an image processing algorithm identifies the number of seeds, cross-validating with a photoelectric sensor. A capacitive material flow sensor is added to the fertilizer channel, measuring flow rate by utilizing the capacitance change caused by material passage, cross-validating with an impact sensor. The controller compares the measurement results of the two sensors in real time, calculating the relative deviation. A deviation less than 15% of the flow rate value is considered normal, and the average value is taken as the final measurement. A deviation exceeding 15% and occurring more than five times consecutively is considered a sensor fault. Based on the historical reliability of each sensor, one output is selected, and a fault alarm is issued to prompt operators to check and maintain the sensor.

[0033] Temperature compensation is applied to the flow sensor because its zero point and sensitivity drift with temperature changes. A temperature sensor is installed near the sensor to monitor the ambient temperature in real time, and the measured values ​​are corrected based on the sensor's temperature characteristic curve. This temperature characteristic curve, obtained through calibration tests at different temperatures, describes the relationship between zero point drift, sensitivity changes, and temperature. The controller obtains the correction coefficient from a table based on the current temperature and calculates the correction value between adjacent temperature points using linear interpolation. This corrects the measured value, eliminating the influence of temperature and improving measurement accuracy.

[0034] The flow measurement value is judged to be reasonable. Based on information such as the remaining amount of material in the material box, the maximum output capacity of the seed and fertilizer dispenser, and the travel speed, the theoretically possible flow range is calculated. When the measured value exceeds the reasonable range, it is judged as abnormal, which may be due to sensor failure or material blockage. The abnormality handling program is triggered, abnormal information is recorded, an alarm is issued, and if necessary, the operation speed is reduced or the operation is stopped to ensure operation safety.

[0035] The output of this step is the fused high-precision travel speed, speed estimation confidence, speed change rate, actual seed application rate, actual fertilizer application rate, flow measurement confidence, and health status indicators of each sensor. This information constitutes the operational status perception dataset, providing accurate and reliable input for subsequent dynamic adjustment and control.

[0036] Step 2: Based on the operation status awareness dataset, test the initial parameters of material characteristics, calculate the target flow rate of dual materials, perform online identification and continuous update of material characteristic parameters, and obtain the material characteristic adaptive modeling result set; Specifically, the following steps are included: Step 2.1: Loading initial parameters for material properties and testing before operation; Based on the material type information input by the operator through the human-machine interface, including seed and fertilizer variety names, database query technology is used to load the initial characteristic parameters of the corresponding materials from the material characteristic database, obtaining the nominal values ​​of parameters such as the flowability coefficient, friction coefficient, bulk density, and particle size distribution of the seeds and fertilizers. The material characteristic database is stored in the controller's non-volatile memory and includes characteristic parameters of common crop seeds and fertilizer varieties. These parameters are derived from standard tests or historical operation data statistics. The flowability coefficient describes the ease with which the material flows in the spreading mechanism. Materials with good flowability have a large coefficient, resulting in uniform and stable flow, while materials with poor flowability have a small coefficient, making them prone to bridging and clogging. The friction coefficient describes the frictional characteristics between the material and the wall of the spreading mechanism, affecting the material's resistance to movement and discharge speed within the mechanism. The bulk density describes the mass-to-volume ratio of the material in its loosely packed state, affecting the conversion relationship between volumetric flow rate and mass flow rate. The particle size distribution describes the statistical characteristics of the material's particle size, affecting the filling and discharge process of the material in the seed and fertilizer metering device.

[0037] Before the formal operation begins, a material characteristic test program is automatically executed to verify whether the database parameters match the actual materials and to correct for batch differences. The test program controls the seed and fertilizer metering device to perform a series of preset test actions, including step response tests and frequency response tests.

[0038] In the step response test, the controller issues a step-change speed command, causing the seed and fertilizer metering motor to suddenly accelerate from a standstill to a set speed, then maintain a constant speed for a period of time, before suddenly decelerating to a standstill. The flow sensor records the material flow rate change curve throughout the process. By analyzing the characteristic parameters of the flow rate curve, such as rise time, steady-state value, overshoot, and settling time, the flow response characteristics of the material can be identified. The rise time reflects the ease with which the material starts flowing; the steady-state value reflects the relationship between flow rate and speed during stable flow, and is related to the bulk density and structural parameters of the seed and fertilizer metering device; the overshoot reflects the inertial characteristics of the material, and is related to particle mass and flowability.

[0039] In frequency response testing, the controller issues sinusoidal speed commands, causing the motor speed of the seed and fertilizer metering device to change periodically according to a sinusoidal law. The test frequency scans from low to high frequencies, and the flow sensor records the amplitude and phase lag of the flow response at each frequency. By analyzing the amplitude-frequency and phase-frequency characteristics of the flow response, the dynamic characteristics of the material flow system can be identified. In the low-frequency range, the material flow rate can track speed changes well, with an amplitude ratio close to 1 and a small phase lag. In the high-frequency range, due to the influence of material inertia and frictional damping, the flow tracking ability decreases, the amplitude ratio decreases, and the phase lag increases. Different materials have different frequency response characteristics. By comparing with standard characteristics, the differences between the actual material and the nominal material in the database can be identified.

[0040] Based on test data, a parameter identification algorithm is used to obtain estimated values ​​of the characteristic parameters of the actual material. Parameter identification employs the least squares method to establish a parameterized mathematical model of the material flow process. The model includes the material characteristic parameters to be identified. The rotational speed during the test is used as the model input, and the measured flow rate is used as the expected output. The model parameters are adjusted to minimize the sum of squared errors between the model output and the actual measured flow rate. The parameter values ​​that minimize the error are then obtained as the identification results. The material flow model is described using a transfer function. The numerator coefficient of the transfer function is related to the material bulk density and the geometric parameters of the seed and fertilizer dispensers, while the denominator coefficient is related to the inertia and damping characteristics of the material flow. By identifying the transfer function coefficients, the material characteristic parameters can be deduced.

[0041] The identified parameters are compared with the nominal parameters in the database to calculate the relative deviation. When the deviation is within a reasonable range, the identified parameters are used to update the database parameters. When the deviation exceeds a reasonable range, the operator is prompted to confirm whether the material type has been selected correctly to avoid parameter errors caused by misselection. The updated parameters serve as the benchmark for material characteristic parameters in this operation, stored in the controller's working memory for real-time calculation, and simultaneously written to the operation log area of ​​non-volatile memory for online updates and long-term database optimization in subsequent steps 2.3. If the deviation between the identified parameters and the nominal parameters in the database exceeds 30%, the system pauses and enters operation mode, prompting the operator to reconfirm the material type selection or check the material quality to avoid uncontrolled application due to incorrect parameters.

[0042] Step 2.2, Dynamic calculation of dual material target flow rate based on speed and operating parameters; Based on the high-precision travel speed and speed change rate fused from the operation status perception dataset, combined with the operation width, target unit area application rate and material characteristic parameters, a speed-flow rate mapping model is adopted to dynamically calculate the target instantaneous application flow rate that seeds and fertilizers should achieve at the current moment, and obtain the target mass flow rate of seeds and the target mass flow rate of fertilizers.

[0043] The basic principle of speed-flow mapping is that the field area traversed by the machinery per unit time is equal to the travel speed multiplied by the operating width. To ensure that the material application rate per unit area reaches the target value, the total amount of material applied per unit time should be equal to the area traversed per unit time multiplied by the target application rate per unit area. In other words, the target application flow rate equals the travel speed multiplied by the operating width multiplied by the target application rate per unit area. For seeds, the target application rate per unit area is expressed in terms of the number of seeds, which needs to be converted to mass flow rate. The calculation of the target mass flow rate for seeds includes: calculating the field area traversed per unit time based on the travel speed and operating width; multiplying this area by the target number of seeds per unit area to obtain the number of seeds to be applied; then converting the number of seeds to mass, i.e., multiplying by the weight of 1000 seeds and dividing by 1000, to obtain the target mass flow rate for seeds. For fertilizers, the target application rate per unit area is directly expressed in terms of mass. The target mass flow rate for fertilizers equals the travel speed multiplied by the operating width multiplied by the target application rate per unit area.

[0044] Based on basic calculations, material characteristic compensation is introduced to correct for differences in the flow characteristics of different materials. For materials with poor flowability, bridging and blockage are prone to occur in the seed and fertilizer metering device, and the actual outflow will be less than the theoretical calculation value. A compensation coefficient needs to be increased, with the target flow rate multiplied by a compensation coefficient greater than 1 for correction. For materials with good flowability, excessively smooth flow may lead to over-discharge, requiring a reduction in the compensation coefficient. The target flow rate multiplied by a compensation coefficient less than 1 for correction. The compensation coefficient is determined by referring to a table based on the material's flowability coefficient. Materials with high flowability coefficients have compensation coefficients close to 1, while materials with low flowability coefficients have compensation coefficients greater than 1. The specific correspondence is obtained through fitting a large amount of experimental data and stored in the controller.

[0045] The purpose of introducing feedforward compensation based on the velocity change rate is to reduce control lag and improve the ability to track rapid velocity changes. The velocity change rate is equal to the current velocity minus the previous velocity, divided by the sampling time interval; a positive value indicates acceleration, and a negative value indicates deceleration. When an upward velocity trend is detected (i.e., the velocity change rate is positive), it is predicted that the velocity will continue to increase in the near future, requiring an increase in the target flow rate. Therefore, a feedforward compensation is added to the currently calculated target flow rate. The magnitude of the compensation is proportional to the velocity change rate. The calculation of the feedforward compensation considers the impact of the velocity change trend on the flow demand. Based on the velocity change rate and the working width, the area change caused by the velocity change per unit time is calculated, multiplied by the target application rate per unit area to obtain the flow change, and then multiplied by a feedforward coefficient for adjustment to obtain the feedforward compensation. The feedforward coefficient is determined based on the system's dynamic characteristics and control cycle, typically between 0.3 and 0.7, and is optimized through simulation and experimentation. When a downward velocity trend is detected (i.e., the velocity change rate is negative), the feedforward compensation is negative, reducing the target flow rate in advance to avoid over-application after the velocity slows down.

[0046] The calculated target flow rate is limited by amplitude and rate of change. When the target flow rate cannot reach the theoretical value due to amplitude or rate of change limitations, the system records the deviation between the actual target flow rate and the theoretical value, accumulating the amount of under- or over-spreading material caused by amplitude limitations. When operating conditions permit, such as entering a stable speed section or when the load on the actuator decreases, the system compensates and adjusts based on the accumulated deviation, appropriately increasing or decreasing the spreading amount in subsequent operations to ensure that the average spreading amount for the entire plot still reaches the target value, avoiding total deviation caused by local amplitude limitations. The upper limit of the target flow rate is set as the maximum output flow rate of the seed and fertilizer metering device at its highest speed, and the lower limit is set as the minimum flow rate to maintain stable flow. When the calculated value exceeds the range, saturation amplitude limitation is applied, and the boundary value is taken as the actual target flow rate. The rate of change limitation of the target flow rate is set as the maximum rate of change that the actuator can track, determined based on the acceleration and deceleration capability of the servo motor and the inertia of the spreading mechanism. When the target flow rate change exceeds the limit between adjacent moments, the slope of the change is limited, so that the target flow rate smoothly transitions to the new target value according to the maximum allowable rate of change, avoiding shocks and oscillations caused by step changes.

[0047] For both seeds and fertilizers, the above calculations are performed separately to obtain the target flow rates for seeds and fertilizers. The ratio of these two target flow rates reflects the application ratio of the two materials. During the calculation process, it is ensured that the calculation of both target flow rates uses completely consistent travel speed and working width data, and uses the speed measurement values ​​at the same time to avoid ratio deviations caused by data asynchrony. The two calculated target flow rates are output simultaneously and enter the next step of the actuator control process.

[0048] Step 2.3: Online updating of material property parameters and continuous optimization of the model; Based on control commands and actual flow data accumulated during operations, recursive least squares method is used to update material characteristic parameters online, obtaining progressively optimized model parameters and improving model accuracy. The difference between online updates and pre-operation testing lies in the fact that pre-operation testing involves specialized stimulus experiments under static conditions, resulting in high-quality data but being time-consuming. Online updates, on the other hand, utilize actual operational data during normal operations. While the data quality is relatively lower due to various factors, it accurately reflects the true operational conditions without incurring additional time costs.

[0049] Before online updates, the operational data needs to be screened to remove data segments unsuitable for parameter identification. Screening criteria include: the absolute value of the rate of change of travel speed is greater than 0.2 km / h / s, indicating sufficient excitation for speed change; high confidence level of flow sensor measurements to ensure data reliability; control commands not being in a saturated state to avoid saturated data contaminating the identification results; and high or medium confidence level of speed estimation to exclude periods of sensor anomalies. Only data that simultaneously meets the above conditions can be used for parameter updates via the recursive least squares algorithm, improving identification accuracy and convergence stability.

[0050] The basic process of recursive least squares is as follows: based on the current parameter estimate and the current control command, calculate the flow output predicted by the model, compare the predicted flow with the actual flow measured by the sensor to obtain the prediction error, calculate the update gain based on the prediction error and the error covariance matrix, use the update gain to correct the parameter estimate to obtain a new parameter estimate, and at the same time update the error covariance matrix to prepare for the next update.

[0051] To ensure the stability and convergence of online updates, a forgetting factor mechanism is introduced, giving higher weight to recent data and gradually decreasing the weight of older data, enabling the algorithm to track the slow time-varying characteristics of parameters. The forgetting factor is set between 0.95 and 0.99; a value close to 1 results in a longer memory time and smoother parameter estimation but slower tracking speed, while a value less than 1 results in a shorter memory time and faster tracking speed but is more susceptible to noise. An appropriate forgetting factor is selected based on the time-varying rate of the actual material characteristics.

[0052] Reasonable constraints are imposed on parameter updates, limiting parameter variations to a physically reasonable range to prevent abnormal data from causing parameter divergence. After each update, the new parameters are checked. If a parameter exceeds a reasonable range, the update is rejected, the previous parameter estimate is maintained, and abnormal events are recorded. If multiple consecutive update anomalies occur, it indicates a possible sensor malfunction or material anomaly, requiring manual inspection.

[0053] The parameters obtained from online updates replace the parameters in the model in real time and are used to calculate the target flow rate for the next time step. This allows the model to continuously reflect reality as the operation progresses, improving calculation accuracy. The update process continues in the background without affecting real-time control. The parameter update cycle can be set to an integer multiple of the control cycle, such as updating parameters once every 10 control cycles, balancing the update frequency and computational burden.

[0054] The output of this step is a set of adaptive modeling results for material properties, including: instantaneous application rate of seed target, instantaneous application rate of fertilizer target, feedforward compensation amount, current material flowability coefficient, friction coefficient, bulk density and other updated characteristic parameters, model prediction accuracy index. This information provides a basis for the precise control of the actuator and the evaluation of control performance.

[0055] Step 3: Based on the adaptive modeling result set of material characteristics, construct the mechanism inverse model to calculate the feedforward control command, calculate the feedback control quantity to obtain the comprehensive control command, and superimpose predictive compensation and dead zone compensation to obtain the comprehensive control command set of the dual-channel actuator; Specifically, the following steps are included: Step 3.1, Dynamic characteristic modeling and inverse model construction of the actuator; Based on the physical structure and working principle of the seed and fertilizer discharge actuator, a mechanism modeling method is used to establish a dynamic transmission relationship model from control commands to material output flow of the actuator, resulting in a positive dynamic model.

[0056] A forward dynamic model was established for the seed and fertilizer dispensing actuators. The seed channel uses an external grooved wheel seed metering device, where the seed dispensing flow rate exhibits a non-linear relationship with rotational speed. A quadratic polynomial is used to describe this relationship, and the coefficients are obtained by fitting actual flow rates measured at different rotational speeds during bench calibration experiments. The fertilizer channel uses a screw conveyor type fertilizer metering device, where the fertilizer dispensing flow rate also exhibits a non-linear relationship with rotational speed. The flow rate-rotational speed function was obtained through bench calibration experiments.

[0057] The dynamic response of the servo motor drive system is described using a first-order inertial element, with the time constant reflecting the response delay from the speed command to the actual speed. A complete forward model from the speed command to the material flow rate is established by integrating the flow characteristics of the actuator and the dynamic characteristics of the motor.

[0058] An inverse model is constructed based on the forward model for feedforward control. The inverse function of the static nonlinear relationship between flow rate and speed is solved to obtain the mapping relationship between speed and flow rate. A first-order lead compensation element is constructed for the first-order inertial element, and the static and dynamic inverse models are cascaded to form a complete inverse model. The differential operations in the inverse model use a first-order backward difference approximation, and the difference results are low-pass filtered to suppress noise.

[0059] Step 3.2, Calculation of feedforward control quantity and prediction and compensation of speed change; Based on the constructed inverse model and the instantaneous seed and fertilizer application rates from the adaptive modeling results set of material properties, a feedforward control strategy is adopted to calculate the feedforward control quantity, thereby obtaining the feedforward speed commands for the seed and fertilizer metering motors. The seed target flow rate is input into the seed channel inverse model, and the steady-state speed is obtained through the static inverse function. Dynamic compensation is then performed through the dynamic inverse model to obtain the feedforward speed command. The feedforward speed command can compensate for the main static nonlinearities and dynamic delays of the system, making the actuator output closer to the target value and reducing the deviation that needs to be corrected by feedback control.

[0060] Additional feedforward compensation based on speed change prediction is introduced. Based on the fused high-precision travel speed and speed change rate information from the operational status perception dataset, the speed change trend for the next few steps is predicted, allowing for advance adjustment of control commands. A linear prediction method is employed, where the predicted speed for the future moment equals the current speed plus the speed change rate multiplied by the prediction time step. The prediction time step is set to half the system response delay, typically 0.1 to 0.2 seconds. The target flow rate is recalculated based on the predicted speed, and the difference between this and the current target flow rate is used as the expected flow rate change. This expected flow rate change is multiplied by a prediction compensation coefficient and then superimposed onto the feedforward control command to achieve proactive adjustment. The prediction compensation coefficient is adaptively adjusted based on the smoothness of the speed change; the coefficient is 0.5 to 0.8 when the standard deviation of the speed change rate is less than a set threshold, and decreases to 0.1 to 0.3 when it exceeds the threshold.

[0061] Saturation limiting is applied to feedforward control commands to confine them within the operating capabilities of the motor and actuator, preventing execution failure or mechanical damage due to exceeding capabilities. The rate of change of feedforward commands is also limited to ensure smooth command changes and avoid mechanical shocks and oscillations caused by step commands.

[0062] Step 3.3, Feedback controller design and flow deviation correction; Based on the deviations between the actual seed and fertilizer application rates in the operational status perception dataset and the target flow rates in the material characteristic adaptive modeling result set, a nonlinear PID feedback control algorithm is used to calculate the feedback control quantity, thereby obtaining the speed adjustment amount for deviation correction. The flow rate deviation equals the target flow rate minus the actual flow rate; a positive deviation indicates that the actual flow rate is too low and the control quantity needs to be increased, while a negative deviation indicates that the actual flow rate is too high and the control quantity needs to be decreased.

[0063] A nonlinear PID controller comprises three components: proportional, integral, and derivative. The proportional component generates the control action based on the current deviation. The control quantity of the proportional component is determined by the product of the current deviation and the proportional coefficient. The larger the proportional coefficient, the stronger the response to deviation. A nonlinear proportional gain is introduced, which adaptively adjusts according to the magnitude of the deviation. When the absolute value of the deviation is greater than the flow deviation switching threshold, a large gain is used to accelerate the response; when the absolute value of the deviation is less than the flow deviation switching threshold, a small gain is used to ensure stability. The gain switching uses a smooth transition function to avoid abrupt changes. The flow deviation switching threshold is preferably set to 10% of the target flow rate.

[0064] The integral stage generates control based on the accumulation of deviation, eliminating steady-state error. The control quantity of the integral stage is calculated based on the cumulative effect of the deviation, using a numerical integration method. This involves multiplying the current deviation by the sampling period and adding it to the integral value from the previous moment to obtain the current integral value. This integral value is then multiplied by the integral coefficient to obtain the integral control quantity. Integral separation and integral limiting mechanisms are introduced. When the absolute value of the deviation exceeds 20% of the target flow rate, integration is paused to avoid integral saturation. When the control quantity reaches saturation, integration is stopped to prevent integral accumulation. The upper limit of the absolute value of the integral is limited to prevent excessive integral values ​​from causing overshoot.

[0065] The differential stage generates control based on the changing trend of the deviation, achieving proactive adjustment. The control quantity of the differential stage is calculated based on the changing trend of the deviation. First, the deviation change rate is calculated using a first-order backward differential method, i.e., the deviation at the current time step is subtracted from the deviation at the previous time step, and then divided by the sampling period. The deviation change rate is then multiplied by the differential coefficient to obtain the differential control quantity. To address the sensitivity of differential operations to noise, the deviation signal is first low-pass filtered before differentiation, or the differential result is filtered to suppress high-frequency noise amplification. An incomplete differential form is introduced, adding a low-pass filter to the differential stage. The dynamic characteristics of the filter are determined by both the differential coefficient and the filtering time constant. A larger filtering time constant results in a stronger filtering effect and better noise suppression. The filtering time constant is typically 5 to 10 times the sampling period.

[0066] The control quantities from the three stages are added together to obtain the PID output. The feedback control quantity equals the proportional quantity plus the integral quantity plus the derivative quantity. The feedback control quantity is then superimposed on the feedforward control quantity to obtain the comprehensive control command. The comprehensive speed command equals the feedforward speed command plus the feedback control quantity.

[0067] Independent PID controllers are designed for the seed and fertilizer channels, respectively, with each channel implementing feedback control based on its flow deviation. The PID parameters for the two channels can differ, and are independently tuned according to the characteristics of their respective actuators and materials. Since the seed and fertilizer channels have different dynamic characteristics and noise levels, using targeted parameters can achieve better control results.

[0068] PID parameter tuning is divided into two stages: offline tuning and online optimization. During offline tuning, the initial parameters are determined using the critical proportional gain method. Overshoot, oscillation period, and settling time are measured through step response tests. Parameters are adjusted based on the response characteristics until the performance targets are met. During online optimization, control performance indicators, including the root mean square deviation and the rate of change of the control quantity, are calculated at fixed intervals. When the root mean square deviation increases for three consecutive cycles, the proportional gain is decreased by 10%. When the absolute value of the steady-state deviation exceeds 5% of the target flow rate, the integral gain is increased by 10%. The parameter adjustment range is limited to ±50% of the initial value.

[0069] Step 3.4, Dead zone compensation and friction compensation of the actuator; Based on the nonlinear characteristics of actuators, such as mechanical backlash and friction, nonlinear compensation technology is employed to add compensation amounts to the control commands, resulting in compensated control commands and improved control accuracy. Mechanical backlash exhibits dead-zone characteristics; when the control command changes within a small range, the actuator does not respond, only acting when the command change exceeds the dead-zone width. Dead-zone causes distortion of small signals, making it impossible for the actuator to track small changes in the target flow rate.

[0070] The dead-zone compensation method involves superimposing a fixed compensation amount in the direction of control command change, with the compensation amount amplitude set to half the dead-zone width. The dead-zone width is determined by slowly changing the control command and observing the critical value of the actuator's response; the difference between the critical values ​​in the positive and negative directions is the dead-zone width.

[0071] Friction compensation employs a segmented model, dividing the actuator's operating state into three stages: static, startup, and operation. In the static state, the compensation is zero. At startup, a static friction compensation torque is superimposed. In the operation state, a dynamic friction compensation torque and a viscous friction compensation torque proportional to velocity are superimposed. Friction model parameters are determined through constant torque loading tests. Steady-state velocities under different torques are recorded, and torque-velocity relationship curves are fitted to obtain the various friction parameters.

[0072] By combining dead zone compensation and friction compensation, two compensation values ​​are superimposed on the original control command to obtain the final control command sent to the servo drive, thereby improving the nonlinear characteristics of the actuator and increasing the flow tracking accuracy.

[0073] This step outputs a dual-channel actuator integrated control instruction set, including: speed control instructions for the seed metering servo motor, speed control instructions for the fertilizer metering servo motor, feedforward control quantities, feedback control quantities, integrated control instructions, predictive compensation quantities, and dead zone friction compensation quantities. These control signals are sent to the servo driver via the CAN bus to drive the actuator to move precisely.

[0074] Step 4: Based on the integrated control instruction set of the dual-channel actuator, establish a coupling model to eliminate interference between channels, calculate the allocation deviation and perform multi-objective optimization, and obtain the coordinated and optimized control instruction set according to the scheduling priority of each channel's saturation state. Specifically, the following steps are included: Step 4.1, Modeling and evaluating the coupling characteristics between channels; Based on the physical structure of the dual-channel system, potential coupling mechanisms are analyzed, a coupling characteristic model is established, and the coupling transfer function is obtained. The physical coupling between the two dispensing channels mainly originates from the shared material bin. Both channels draw material from the same bin, and the discharge from one channel causes changes in material flow and pressure within the bin, affecting the discharge process of the other channel. When one channel suddenly increases its flow rate, it draws more material from the bin, accelerating the flow of material from the bin into that channel. This may slow down the material flow in the other channel, causing a short-term decrease in the flow rate of the other channel.

[0075] A coupling model is established, defining the coupling transfer function of channel 1 to channel 2, describing the dynamic relationship between changes in the control input of channel 1 and changes in the output of channel 2. Using an experimental identification method, the control input of channel 2 is kept constant, and a step change in the control input is applied to channel 1. The change in the output flow rate of channel 2 is observed, and the coupling response curve is recorded. The coupling transfer function is obtained by fitting the coupling response curve, typically described by a first-order inertial element with a pure delay. The transfer function parameters include coupling gain, coupling time constant, and coupling delay. The coupling transfer function of channel 2 to channel 1 is identified using the same method, resulting in a complete bidirectional coupling model.

[0076] The coupling degree index is calculated to assess the strength of the mutual influence between two channels. The coupling degree is equal to the ratio of the coupling gain to the main channel gain. A low coupling degree indicates a weak coupling effect, which can be approximately independently controlled. A high coupling degree indicates a strong coupling effect, which requires decoupling control. Based on the coupling degree calculated from the identification results, the degree of coupling influence is judged. When the coupling degree exceeds a preset coupling degree threshold, the decoupling control mode is activated. A compensation strategy is used to eliminate inter-channel interference and improve control accuracy.

[0077] Step 4.2, Design of feedforward decoupling controller and calculation of coupling compensation; Based on the coupling model, a feedforward decoupling control strategy is adopted to calculate the decoupling compensation amount and obtain the coupling compensation control command for each channel. The basic idea of ​​decoupling control is that when control is applied to one channel, compensation control is applied to another channel at the same time. The role of the compensation amount is to cancel the coupling effect of the former channel on the latter channel, so that the output of each channel is only affected by the control of its own channel and is not disturbed by the other channel.

[0078] For each channel, its output is influenced by both the control command of that channel and the coupling effect of the other channel. The decoupling compensation is calculated by inputting the control command of the other channel into the coupling transfer function to obtain the coupling interference experienced by that channel; the compensation is the inverse of this interference. Since the coupling transfer function contains dynamic elements, the compensation is calculated using a digital filter, with filter parameters set according to the identified coupling model. The decoupling compensation of each channel is then superimposed onto the original independent control command to obtain the decoupled control command, thereby eliminating mutual interference between channels.

[0079] Decoupling control can effectively reduce mutual interference between channels, but complete decoupling requires an accurate coupling model. In practice, model errors and time-varying parameters can lead to incomplete decoupling, and residual coupling can be further eliminated through feedback control.

[0080] Step 4.3, Flow ratio deviation detection and coordinated optimization control; Based on the actual seed and fertilizer application rates in the operational status-aware dataset, the actual ratio is calculated and compared with the target ratio in the material characteristic adaptive modeling result set to obtain the ratio deviation. A coordinated optimization strategy is then used to correct the ratio. The actual ratio equals the ratio of the actual seed application rate to the actual fertilizer application rate, and the target ratio equals the ratio of the target seed application rate to the target fertilizer application rate. The ratio deviation equals the actual ratio minus the target ratio. A positive deviation indicates that there is relatively too much seed or too little fertilizer, while a negative deviation indicates that there is relatively too little seed or too much fertilizer.

[0081] When the absolute value of the ratio deviation exceeds the ratio deviation trigger threshold, the ratio coordination control mode is activated. The ratio deviation trigger threshold is preferably set at 5% of the target ratio. Priority is given to correcting the ratio relationship, and the requirements for absolute flow deviation of each channel are temporarily relaxed. The coordination control strategy is to determine which channel has a relatively larger flow deviation, apply stronger control to the channel with the larger deviation to accelerate correction, and appropriately relax control on the channel with the smaller deviation. Synchronous optimization is then performed after the channel with the larger deviation recovers.

[0082] A multi-objective optimization function is established, with objectives including minimizing the flow deviation of channel 1, minimizing the flow deviation of channel 2, minimizing the flow ratio deviation, and optimizing the smoothness of control command changes. The overall objective function is obtained by weighting and summing the objectives according to their respective weight coefficients. When the flow ratio deviation is large, the weight of the flow ratio term is increased; when the flow ratio deviation is small, the weight of the flow ratio term is decreased, emphasizing the absolute accuracy of each channel. A constrained optimization algorithm is used to solve for the optimal control quantity. The constraints include upper and lower limits of the control quantity, the rate of change of the control quantity, and the physical realizability of the flow.

[0083] The optimization solution employs an iterative algorithm. At the current operating point, the gradient of the objective function and constraints are calculated to determine the optimization direction and step size. The control input is then updated, and the iterations are repeated until the objective function converges or the preset number of iterations is reached. Considering the time requirements of real-time control, the number of iterations is limited to five to ensure that the calculation is completed within the control cycle.

[0084] The optimized control quantity is used to modify the independent control commands of each channel in the integrated control command set of the dual-channel actuator. The modified control commands take into account both decoupling compensation and matching coordination to achieve optimal coordinated control of the dual channels.

[0085] Step 4.4, Dynamic scheduling and saturation handling of channel priorities; Based on the operating status and control saturation of each channel, a dynamic priority scheduling strategy is adopted to rationally allocate control resources and obtain an optimized control allocation scheme. When the control command of a channel reaches the saturation limit and the actuator is already working at its limit and cannot be further adjusted, the flow deviation of that channel cannot be eliminated by controlling that channel, and it is necessary to adjust another channel to maintain the matching relationship.

[0086] Channel priorities are defined and dynamically calculated based on factors such as the magnitude of flow deviation, duration of deviation, and control saturation level. Channels with larger deviations, longer durations, and those that are not yet saturated have higher priority, while channels with smaller deviations and those that are already saturated have lower priority. Priority is given to ensuring the flow accuracy of high-priority channels, while lower-priority channels, to meet the requirements of allocation and coordination, may sacrifice accuracy appropriately.

[0087] When saturation is detected in one channel, the control input for that channel is frozen at the saturation value. The matching ratio is corrected by adjusting the control input of the other channel. The target flow rate of the other channel is calculated by back-calculating the actual flow rate of the saturated channel and the target matching ratio, so that the actual matching ratio equals the target matching ratio. When the flow deviation of the saturated channel decreases and the control command leaves the saturation limit, the frozen state of that channel is released, and the two channels resume normal coordinated control mode, gradually adjusting to their respective target flow rates. Although this strategy causes the absolute flow rates of both channels to deviate from the original target values, it ensures the matching relationship and achieves a suboptimal solution within the limitations of the actuator's capabilities.

[0088] When both channels reach saturation, it indicates that the target flow rate exceeds the system's capacity, triggering an over-limit alarm. This alerts operators to reduce the operating speed or the amount of target application to prevent prolonged operation beyond limits and potential equipment damage. The system automatically reduces the target flow rate proportionally to within the system's capacity, maintaining the ratio, and continues operation until conditions allow, at which point it returns to the normal target value.

[0089] This step outputs a coordinated and optimized control command set, which comprehensively considers decoupling compensation and proportion coordination, and is directly sent to the actuators of each channel. Simultaneously, it outputs monitoring information such as proportion deviation, channel priority, and saturation status for system status display and anomaly alarms, ensuring precise and synchronized application of the two materials.

[0090] Step 5: Based on the coordinated and optimized control instruction set, record the data of the entire operation process and calculate the quality indicators, establish a control strategy model to optimize control decisions, establish a multi-condition parameter library, and obtain a set of continuous improvement and optimization results. Specifically, the following steps are included: Step 5.1, Data storage and quality index calculation for the job; The operation status perception dataset, material characteristic adaptive modeling result set, dual-channel actuator integrated control instruction set, coordinated and optimized control instruction set, and monitoring information such as ratio deviation, channel priority, and saturation status collected during the operation are stored in the controller memory in time series format to form a complete operation data log.

[0091] After the operation is completed, the recorded data are used to calculate quality indicators to evaluate the control performance of the operation and obtain quantified performance evaluation indicators. The average deviation of seed flow rate is calculated, which is equal to the average of the flow rate deviations at all times, reflecting systematic errors. The maximum deviation of seed flow rate is calculated, which is equal to the maximum absolute value of the flow rate deviations at all times, reflecting the control effect under the worst operating conditions. The standard deviation of seed flow rate deviation is calculated, reflecting the degree of flow rate fluctuation; a small standard deviation indicates stable control, while a large standard deviation indicates large control fluctuations. The same method is used to calculate various indicators of fertilizer flow rate. These quality indicators are used to evaluate the operation effect and for subsequent parameter optimization.

[0092] Calculate the proportion deviation index, including the average, maximum, and standard deviation of the proportion deviation, to evaluate the effectiveness of the coordinated control of the two materials. Calculate the application uniformity using the coefficient of variation, which equals the standard deviation divided by the average; a smaller coefficient of variation indicates better uniformity. Calculate the smoothness of the control commands, using the standard deviation of the difference between control commands at adjacent time points; good smoothness indicates stable control without drastic fluctuations.

[0093] Each indicator is compared with the preset quality requirements to determine whether it meets the standards. The standards include: the average absolute value of the seed flow rate deviation is less than 3% of the target flow rate, the maximum deviation is less than 8% of the target flow rate, and the standard deviation is less than 5% of the target flow rate; the average absolute value of the fertilizer flow rate deviation is less than 3% of the target flow rate, the maximum deviation is less than 8% of the target flow rate, and the standard deviation is less than 5% of the target flow rate; the average absolute value of the ratio deviation is less than 2% of the target ratio, and the maximum ratio deviation is less than 5% of the target ratio; the coefficient of variation for application uniformity is less than 10%; and the standard deviation of control command smoothness is less than 15% of the maximum control amount. The reasons for non-compliance are analyzed in detail to provide direction for parameter optimization. An operation quality report is generated, including indicator values, compliance status, and data curves and charts, presented to operators in an intuitive format to facilitate operation effect evaluation and problem diagnosis.

[0094] Step 5.2, frequency response analysis and dynamic performance evaluation of the control system; Based on velocity variation and flow response data from the operational data, frequency domain analysis is employed to evaluate the dynamic characteristics of the control system and obtain its frequency response characteristic curve. Through frequency domain analysis, the main frequency components of velocity variation are extracted, and the flow rate's ability to track velocity variations at different frequencies and its response delay are evaluated. The system's bandwidth frequency and resonant frequency are then identified. The bandwidth frequency reflects the highest velocity variation frequency that the system can track, while the resonant frequency indicates the frequency points where the system is prone to oscillations, which need to be suppressed through parameter adjustments.

[0095] Identify the system's bandwidth frequency, which is the frequency at which the amplitude-to-frequency ratio drops to 0.707. A larger bandwidth indicates that the system can track higher frequencies of velocity changes and has better dynamic performance. Identify the system's resonant frequency, which is the frequency at which the amplitude-to-frequency characteristic shows a peak. Resonance indicates that the system is prone to oscillations near this frequency, and it is necessary to adjust the control parameters to increase damping and suppress resonance.

[0096] By comparing the frequency response characteristics of different operations, the stability and consistency of system performance can be analyzed, performance degradation trends can be identified, and a basis for preventive maintenance can be provided. By comparing the frequency response under different control parameters, the effectiveness of parameter adjustments can be evaluated, providing a quantitative basis for parameter optimization.

[0097] Step 5.3, control strategy learning and optimization based on data-driven model; Based on a large amount of accumulated historical operational data, machine learning methods are used to establish a data-driven control strategy model, resulting in optimized control decisions. The historical data is categorized according to operational conditions, including different crop types, different material varieties, different field conditions, and different operational speed ranges, with each category corresponding to a data subset.

[0098] For each subset of data, a neural network model is built. The network input includes the current speed, rate of change of speed, target flow rate, actual flow rate, and historical flow deviation sequence. The network output is the optimal control variable. A multilayer perceptron structure is adopted, containing an input layer, hidden layers, and an output layer. The number of neurons in the hidden layer is determined according to the problem complexity, typically 2 to 3 times the input dimension. Supervised learning is used to train the network. The training samples are segments of historical data where control performance is good. The corresponding input states are used as sample inputs, and the actual control variable used is used as the label output. The network weights are adjusted through the backpropagation algorithm to make the network output approximate the label value.

[0099] The training process employs the stochastic gradient descent optimization algorithm, sets appropriate learning rates and iteration counts, introduces regularization terms to prevent overfitting, and uses cross-validation to evaluate the model's generalization performance. The dataset is divided into training and test sets, the model is trained using the training set, and its performance is evaluated using the test set. The model with the smallest test error is selected as the final model.

[0100] Considering the limitations of the controller's computing resources, the neural network model adopts a lightweight design, with the number of hidden layer neurons controlled between 20 and 50. Model training is completed on a host computer or in the cloud. The trained network weight parameters are downloaded to the controller to perform inference calculations. The inference process only requires simple matrix multiplication and activation function operations, and the time for a single inference is controlled in the millisecond range, meeting the requirements for real-time control.

[0101] A trained neural network model is embedded into the control system and runs in parallel with traditional model control. The outputs of the two control systems are then weighted and fused. Initially, the neural network weights are small, relying mainly on traditional control. As the neural network model is continuously trained and validated, its weights are gradually increased, ultimately achieving intelligent control. The neural network model can learn complex nonlinear mapping relationships and implicit control laws, exhibiting a stronger adaptability to characteristics that are difficult for traditional models to describe.

[0102] An offline incremental learning mechanism is introduced. After every ten assignments, the newly collected data is uploaded to the host computer or cloud to trigger incremental model training. After training is completed, simulation verification confirms performance improvement and the absence of anomalies. Then, the updated model parameters are downloaded to the controller to replace the original model. The entire update process is performed during non-assignment periods to ensure control stability during assignments.

[0103] Step 5.4: Automatic tuning of control parameters and establishment of a multi-condition parameter library; Based on operational data analysis and model optimization results, an automatic parameter tuning algorithm is employed to optimize PID controller parameters and other key control parameters, yielding optimal parameter configurations for various operating conditions. Parameter tuning utilizes an optimization search method, defining a parameter optimization objective function. The objective is a comprehensive evaluation of control performance indicators, including a weighted sum of multiple indicators such as average deviation, maximum deviation, standard deviation, and response time. Weighting coefficients reflect the importance of each indicator.

[0104] The search is performed using intelligent optimization algorithms, such as genetic algorithms or particle swarm optimization, within the parameter space. Iterative optimization avoids getting trapped in local optima. The fitness of each set of parameters is obtained through simulation evaluation during the search process, and parameter optimization can be completed without actual operation.

[0105] The optimization process employs simulation evaluation, using historical operation speed data as input, and simulated control with the parameters to be optimized. The simulation control effect is calculated as the fitness of the set of parameters. Parameter performance can be evaluated without actual operation, thus improving optimization efficiency.

[0106] For different operating conditions, parameters are optimized separately, and a multi-condition parameter library is established. The parameter library is organized according to dimensions such as crop type, material variety, operating speed range, and field conditions, with a set of optimized parameters corresponding to each operating condition. Before actual operation, the operator selects the current operating condition through the human-machine interface, and the system automatically calls the corresponding parameter configuration from the parameter library. This enables rapid adaptation to different operating needs without the need for manual parameter adjustment, reducing the professional skill requirements of the operator and improving the system's ease of use.

[0107] The parameter library is continuously expanded and updated as operational data accumulates. New operational conditions generate new data, triggering new parameter optimizations. The optimization results are added to the parameter library, and old parameters are periodically re-optimized and updated to keep the parameter library up-to-date.

[0108] The output of this step is a set of continuous improvement and optimization results, including: the optimized set of control parameters, the trained neural network control model, the multi-condition parameter library, the operation quality assessment report, and performance improvement suggestions. These results support the continuous optimization and improvement of control performance, realize intelligent self-learning and self-adaptation, and provide better control strategies and parameter configurations for subsequent operations.

[0109] In some embodiments, due to the complex and variable nature of actual field operations, a single control strategy is unlikely to achieve optimal performance in all situations. A multi-model switching control technique based on operational condition identification can be employed. The aim is to automatically select the most suitable control model and parameters based on real-time operational conditions, thereby improving control adaptability and robustness. Specifically, an operational condition feature library is established, containing speed change pattern characteristics, material property characteristics, and field environment characteristics for different operational conditions. Cluster analysis is used to divide historical operational conditions into several typical categories, each corresponding to a specific operational condition mode. For each operational condition mode, a dedicated control model and parameter configuration are designed to optimize performance under that condition. During actual operations, characteristic parameters of the current operation are extracted in real time, including speed fluctuation frequency, speed change amplitude, and flow response lag time. The similarity between the current features and each typical operational condition is calculated, and the current operational condition category is identified using nearest neighbor or fuzzy matching methods. Based on the operational condition identification results, the corresponding control model is selected from the control model library, and control is switched to that model, achieving dynamic switching of the control mode. To avoid the shock of instantaneous switching, a gradual switching strategy is adopted. During the switching process, the outputs of the two models are weighted and fused according to weight coefficients. The weights smoothly transition from 1 in the old model to 1 in the new model, with the transition time set to several control cycles to ensure a smooth and abrupt switching. Through multi-model switching, the system can adopt the optimal strategy for different operating conditions, exhibiting stronger adaptability and higher overall performance compared to a single model.

[0110] A dynamic adjustment system for precise application of dual materials based on travel speed is provided to execute the aforementioned dynamic adjustment method for precise application of dual materials based on travel speed, such as... Figure 3 As shown, it includes: The operation status sensing module is used to collect raw data from three speed sensors and two material flow sensors and preprocess it to obtain the operation status sensing dataset. The material property adaptive modeling module, based on the operation status perception dataset, tests the initial parameters of material properties, calculates the target flow of dual materials, performs online identification and continuous updates of material property parameters, and obtains the material property adaptive modeling result set. The dual-channel actuator control module constructs an inverse model of the mechanism based on the adaptive modeling result set of material characteristics, calculates feedforward control instructions, calculates feedback control quantities to obtain comprehensive control instructions, and superimposes predictive compensation and dead zone compensation to obtain a comprehensive control instruction set for the dual-channel actuator. The coordinated optimization control module, based on the integrated control instruction set of the dual-channel actuator, establishes a coupling model to eliminate interference between channels, calculates the ratio deviation and performs multi-objective optimization, and obtains the coordinated optimized control instruction set according to the scheduling priority of each channel's saturation state. The continuous improvement and optimization module records data from the entire operation process and calculates quality indicators based on the coordinated and optimized control instruction set. It establishes a control strategy model to optimize control decisions, builds a multi-condition parameter library, and obtains a continuous improvement and optimization result set.

[0111] In one embodiment of the present invention, a specific example is provided: This invention focuses on the application of integrated sowing and fertilization of summer maize in the North China Plain. The operation area is 120 mu (approximately 8 hectares), the terrain is flat but with some local undulations, the soil type is loam, and a 12-row integrated sowing and fertilization machine pulled by a large tractor is used. The operating width is 9 meters and the target operating speed is 10 kilometers per hour.

[0112] Before operation, the parameters were set via the human-machine interface. The corn variety was Zhengdan 958, the target planting density was 4500 seeds per mu (approximately 0.067 hectares), coated seeds with a thousand-seed weight of 350 grams were used, the target fertilizer application rate was 40 kg of compound fertilizer per mu (approximately 0.067 hectares), and slow-release compound fertilizer with a bulk density of 0.85 g / cm³ was used. The seed-fertilizer ratio was 12.6 kg of fertilizer per kg of seeds. The system calculated the target application rate per unit area based on the set parameters: 6.75 corn seeds per square meter and 60 grams of fertilizer per square meter.

[0113] To verify the system's performance, three-day field tests were conducted in the eastern and western areas of the site. During the tests, a high-precision data acquisition system was deployed to simultaneously record key parameters such as three speed sensors, dual-channel flow sensors, and control commands at a frequency of 100Hz. The total test area was 15 acres, covering various typical operating conditions such as flat ground, gentle slopes, and curves.

[0114] During the operation, the system collects operational status sensing data in real time. Table 1 shows examples of raw sensor data from five consecutive time points in the eastern plains region. Table 1: Example of a work status awareness dataset (Eastern Plains Region).

[0115] Table 2 shows examples of control calculation and execution data from five consecutive time points in the gentle western slope area: Table 2: Example of the coordinated and optimized control instruction set (western gentle slope area).

[0116] The two sets of data show that the system can effectively fuse multi-source velocity information under different terrain conditions, and the confidence level of velocity estimation remains at a high level. When the travel speed varies from 8.2 km / h to 11.5 km / h, the system can quickly adjust the target flow rate and control commands. The actual flow rates of seeds and fertilizers closely track the target flow rate, and the actual ratio of the two materials remains stable at around the target ratio of 0.396, with the ratio deviation controlled within 0.001. This verifies the system's precise control capability and dual-channel coordination effect under complex working conditions.

[0117] After the operation was completed, statistical analysis showed that the average sowing density across the entire plot was 4485 seeds per mu (approximately 0.33 hectares), with a coefficient of variation of 6.8, indicating uniform and consistent seedling emergence. The average fertilizer application rate across the entire plot was 39.8 kg per mu (approximately 0.5 kg per hectare), with a uniformity coefficient of variation of 5.2, indicating even fertilizer distribution. The average actual seed-fertilizer ratio was 12.7, with a deviation of 0.8 from the target ratio, showing good stability. Compared to the control plot using the traditional fixed-rate broadcasting method, this invention improved sowing uniformity, fertilizer uniformity, and corn seedling emergence uniformity. Later yield measurements showed a significant increase in yield, verifying the practical application value of this invention.

[0118] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A dynamic adjustment method for precise application of dual materials based on travel speed, characterized in that, Includes the following steps: Raw data from three speed sensors and a dual-channel material flow sensor are collected and preprocessed to obtain an operational status perception dataset. Based on the operational status awareness dataset, test the initial parameters of material characteristics, calculate the target flow rate of dual materials, perform online identification and continuous update of material characteristic parameters, and obtain the material characteristic adaptive modeling result set; Based on the adaptive modeling result set of material properties, the mechanism inverse model is constructed to calculate the feedforward control command, the feedback control quantity is calculated to obtain the comprehensive control command, and the predictive compensation and dead zone compensation are superimposed to obtain the comprehensive control command set of the dual-channel actuator. Based on the integrated control instruction set of the dual-channel actuator, a coupling model is established to eliminate interference between channels, the ratio deviation is calculated and multi-objective optimization is performed, and the coordinated and optimized control instruction set is obtained according to the scheduling priority of each channel's saturation state. Based on the coordinated and optimized control instruction set, the data of the entire operation process is recorded and the quality indicators are calculated. A control strategy model is established to optimize control decisions, a multi-condition parameter library is established, and a set of continuous improvement and optimization results is obtained.

2. The method for dynamic adjustment of precise dual-material application based on travel speed according to claim 1, characterized in that, The steps for obtaining the job status awareness dataset include: The first speed signal is obtained based on the wheel speed sensor. When the difference between the wheel acceleration and the overall machine acceleration exceeds the preset slippage threshold, the confidence weight of the wheel speed sensor is reduced. The speed obtained from the second speed signal based on the GPS positioning module is verified by numerically differentiating the position coordinates and comparing it with the speed directly output by the GPS. The third velocity information is obtained by integrating the acceleration signal using an inertial measurement unit (IMU). The federated Kalman filter algorithm is used to fuse the three velocity information. The bottom layer consists of three local filters that process the three velocities respectively, and the top layer main filter integrates the results of the local filters to obtain the fused velocity estimate and velocity confidence.

3. The method for dynamic adjustment of precise dual-material application based on travel speed according to claim 2, characterized in that, The federated Kalman filter algorithm includes: For the local filter of GPS velocity, the observation noise covariance is dynamically adjusted according to the accuracy factor value output by the GPS module; The main filter performs weighted fusion based on the velocity estimates and error covariance of the local filter outputs, with the weighting coefficients calculated based on the information matrix of each local estimate.

4. The method for dynamic adjustment of precise application of dual materials based on travel speed according to claim 1, characterized in that, The initial parameters of the test material characteristics, the calculation of the target flow rates of the two materials, and the online identification and continuous updating of the material characteristic parameters specifically include: Before the operation begins, perform material characteristic tests and control the seed and fertilizer metering device to perform step response and frequency response tests to identify material flow response characteristics. The least squares method is used to identify the estimated values ​​of the characteristic parameters of the actual material based on the test data; Based on the travel speed, operating width, target unit area application rate and material characteristic parameters, the speed-flow mapping model is used to dynamically calculate the target instantaneous application flow rate. Introduce feedforward compensation for velocity change rate; when a velocity change trend is detected, add feedforward compensation amount to the target flow rate. The recursive least squares method is used for online updating of material property parameters, and a forgetting factor mechanism is introduced.

5. The method for dynamic adjustment of precise application of dual materials based on travel speed according to claim 1, characterized in that, The constructed mechanism uses an inverse model to calculate feedforward control commands, calculates feedback control quantities to obtain comprehensive control commands, and superimposes predictive compensation and dead-zone compensation, including: A dynamic transmission relationship model of the actuator from control commands to material output flow rate is established, and a nonlinear relationship model between seed and fertilizer discharge flow rate and rotation speed is established as a positive dynamic model. Based on the forward dynamic model, an inverse model is constructed to solve the inverse function of the static nonlinear relationship between flow rate and rotational speed, and a first-order advance compensation element is constructed for the first-order inertial element. The target flow rate is input into the inverse model to obtain the feedforward control command; additional feedforward compensation is introduced by the prediction of speed change, and the expected flow rate change is multiplied by the prediction compensation coefficient and then superimposed on the feedforward control command. Based on the deviation between the actual flow rate and the target flow rate, a nonlinear PID feedback control algorithm is used to calculate the feedback control quantity, and a nonlinear proportional gain is introduced to adaptively adjust according to the magnitude of the deviation. Integral separation and integral limiting mechanisms are introduced. Integration is paused when the absolute value of the deviation exceeds the preset integral separation threshold, and integration is stopped when the control quantity reaches saturation. Dead zone compensation and friction compensation torque are superimposed on the control commands.

6. The method for dynamic adjustment of precise dual-material application based on travel speed according to claim 5, characterized in that, The construction of the inverse model based on the forward dynamic model includes: A quadratic polynomial is used to describe the relationship between flow rate and rotational speed. The coefficients are obtained by fitting the actual flow rate measured at different rotational speeds through bench calibration tests. Solving the inverse function of the nonlinear relationship between flow rate and rotational speed yields the mapping relationship between rotational speed and flow rate; The dynamic response of the servo motor drive system is described by a first-order inertial element, and a first-order lead compensation element is constructed as the dynamic inverse model. The static inverse model and the dynamic inverse model are connected in series to form a complete inverse model.

7. The method for dynamic adjustment of precise dual-material application based on travel speed according to claim 1, characterized in that, The establishment of a coupling model to eliminate inter-channel interference, calculation of allocation deviation and multi-objective optimization, and scheduling priority based on the saturation state of each channel include: A coupling characteristic model is established, and an experimental identification method is used to fix the control quantity of one channel and apply a step change control quantity to another channel. The coupling transfer function is then obtained by fitting. Calculate the coupling degree index, and activate the decoupling control mode when the coupling degree exceeds the preset coupling degree threshold; A feedforward decoupling control strategy is adopted to calculate the decoupling compensation amount. The control command of the other channel is input into the coupling transfer function to obtain the coupling interference amount. The compensation amount is the inverse of the interference amount. The allocation deviation is calculated based on the actual flow of the two channels. When the allocation deviation exceeds the preset allocation deviation start threshold, the allocation coordination control mode is activated. A multi-objective optimization function is established, and the objectives are summed with weighted coefficients. The optimal control quantity is then solved using a constrained optimization algorithm. Channel priority is defined and dynamically calculated based on flow deviation, deviation duration, and control saturation level. When saturation is detected in one channel, the ratio is corrected by adjusting the control quantity of another channel.

8. The method for dynamic adjustment of precise dual-material application based on travel speed according to claim 7, characterized in that, The establishment of the multi-objective optimization function includes: The objectives of the multi-objective optimization function include minimizing two flow deviations: minimizing flow ratio deviation and optimizing the smoothness of control command changes; the weight coefficients of each objective are dynamically adjusted according to the magnitude of the ratio deviation. Constraints include upper and lower limits of control variables and control variable rate of change constraints. When one channel is saturated, the target flow rate of the other channel is calculated by back-calculating the actual flow rate of the saturated channel and the target ratio.

9. The method for dynamic adjustment of precise application of dual materials based on travel speed according to claim 1, characterized in that, The process involves recording data throughout the entire operation and calculating quality indicators, establishing a control strategy model for control decision optimization, and building a multi-condition parameter library, including: The data collected during the operation is stored in a time-series format to form an operation data log; After the operation is completed, calculate the average deviation, maximum deviation, and standard deviation of the flow rate, and calculate the average, maximum, and standard deviation of the proportion deviation. The dynamic characteristics of the control system are evaluated using frequency domain analysis based on operational data, and the bandwidth frequency and resonant frequency of the system are identified. A data-driven control strategy model is established using machine learning methods. A neural network model is built, with the network input including the current speed, the rate of change of speed, the target flow rate, the actual flow rate, and the historical sequence of flow deviation. The network output is the optimal control variable. The control parameters are optimized using an automatic parameter tuning algorithm, and a multi-condition parameter library is established for different operating conditions.

10. A dynamic adjustment system for precise application of dual materials based on travel speed, characterized in that, A method for implementing a dynamic adjustment method for precise application of dual materials based on travel speed as described in any one of claims 1-9 includes: The operation status sensing module is used to collect raw data from three speed sensors and two material flow sensors and preprocess it to obtain the operation status sensing dataset. The material property adaptive modeling module, based on the operation status perception dataset, tests the initial parameters of material properties, calculates the target flow of dual materials, performs online identification and continuous updates of material property parameters, and obtains the material property adaptive modeling result set. The dual-channel actuator control module constructs an inverse model of the mechanism based on the adaptive modeling result set of material characteristics, calculates feedforward control instructions, calculates feedback control quantities to obtain comprehensive control instructions, and superimposes predictive compensation and dead zone compensation to obtain a comprehensive control instruction set for the dual-channel actuator. The coordinated optimization control module, based on the integrated control instruction set of the dual-channel actuator, establishes a coupling model to eliminate interference between channels, calculates the ratio deviation and performs multi-objective optimization, and obtains the coordinated optimized control instruction set according to the scheduling priority of each channel's saturation state. The continuous improvement and optimization module records data from the entire operation process and calculates quality indicators based on the coordinated and optimized control instruction set. It establishes a control strategy model to optimize control decisions, builds a multi-condition parameter library, and obtains a continuous improvement and optimization result set.

Images