A sowing unit depth stability active adjustment system and method based on a prediction model
The active adjustment system for seed depth stability, which utilizes predictive and dynamic models, solves the problem of unstable seeding depth in seeders operating at high speeds, achieving stable adjustment of seeding depth and improving seeding quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA AGRI UNIV
- Filing Date
- 2026-02-05
- Publication Date
- 2026-06-05
AI Technical Summary
Existing seeders struggle to adapt to variations in soil hardness during high-speed operation, resulting in inconsistent sowing depths and even "flying seeds." Furthermore, traditional control systems suffer from sluggish response, limited control dimensions, and a lack of environmental awareness, making it difficult to achieve stable adjustment of sowing depth.
An active adjustment system for seeding depth stability based on a predictive model is adopted. The motion parameters of the seeding unit are detected by an inertial measurement unit, a pressure sensor and an angle sensor. Combined with machine-soil coupling dynamic modeling, the seeding depth is predicted in real time and the hydraulic cylinder thrust is adjusted to achieve stable control of the seeding depth.
It achieves stable active adjustment of sowing depth under high-speed operation conditions, reduces sowing depth fluctuations, improves sowing quality and yield, adapts to changes in soil hardness, and avoids severe vibration of individual seeds and the phenomenon of "flying seeds".
Smart Images
Figure CN122139520A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent control and precision seeding technology for agricultural machinery, and in particular to an active adjustment system and method for seed depth stability based on a predictive model. Background Technology
[0002] Sowing depth is a key agronomic indicator affecting crop emergence quality, uniformity of crop population, and final yield. With the development of large-scale agriculture, the operating speed of seeders has been continuously increasing (from the traditional 6-8 km / h to 10-15 km / h), which has brought great challenges to sowing depth control.
[0003] In the existing technology, the mainstream seeding depth control schemes are mainly divided into two categories: (1) Mechanical passive depth limiting: using strong springs or airbags to provide constant downward pressure. This method cannot adapt to the spatial variability of soil hardness, and under high-speed operation, the high-frequency excitation of the ground surface (such as soil clods, stubble, unevenness) will cause the seeding unit to vibrate violently (bounce), resulting in the seeding depth fluctuating, and even the phenomenon of "flying seeds". (2) Feedback-based electro-hydraulic active control: such as the DeltaForce system of PrecisionPlanting in the United States, which monitors the pressure of the depth limiting wheel and uses PID algorithm to adjust the pressure of the hydraulic cylinder. However, such systems have inherent limitations: response lag: relying on feedback adjustment after "error occurs", the response delay of the hydraulic system makes it difficult to cope with transient impacts; single control dimension: only adjusting the downward pressure (i.e., adjusting the stiffness), ignoring the adjustment of the system damping characteristics. In dynamics, the lack of damping makes it difficult to dissipate vibrational energy, and the system is prone to continuous oscillation; lack of environmental perception makes it impossible to distinguish whether the force is increased due to "hard soil" or "rising terrain", and the control strategy is not intelligent enough.
[0004] Therefore, there is an urgent need for an active control system that can start from the essence of dynamic stability, has advanced predictive capabilities, and can simultaneously adjust stiffness and damping. Summary of the Invention
[0005] Therefore, this invention provides an active adjustment system and method for the stability of seeding monomer depth based on a prediction model.
[0006] To achieve the objectives of this invention, the following technical solution is adopted: An active adjustment system for seeding depth stability based on a predictive model is disclosed for controlling the seeding depth of a precision seeder. The system includes a detection device, a control device, and an execution device. The detection device detects the motion parameters of the seeding unit. The control device performs machine-soil coupling dynamic modeling based on the parameters detected by the detection device, predicts the seeding depth of the seeding unit based on the modeling results, and controls the execution device to operate based on the predicted seeding depth. The execution device drives the seeding unit under the control of the control device to ensure that the seeding depth meets a predetermined depth.
[0007] The aforementioned adjustment system includes a detection device comprising an inertial measurement unit, a pressure sensor, a counter-pressure sensor, and an angle sensor.
[0008] The aforementioned adjustment system includes an actuator comprising a pressing cylinder and a lowering cylinder.
[0009] The aforementioned adjustment system, wherein: the control device is used to equate the vertical motion of the seeding unit to a single-degree-of-freedom system, and establish the machine-soil coupled dynamic equations:
[0010] Where: M represents the equivalent mass of the seeding unit. This represents the vertical vibration acceleration of the seeding unit. For inertia, C sys K represents the inherent mechanical damping of the seeding unit suspension system. sys This indicates the inherent mechanical stiffness of the seeding unit suspension system. F represents the vertical velocity of the seed unit, z(k) represents the vertical displacement of the seed unit relative to its equilibrium position, and F soil (k) represents the combined reaction force of the soil on the seed unit, F hydro (k) represents the active thrust applied by the hydraulic cylinder to the seeding unit, F down (k) represents the external downforce.
[0011] The aforementioned adjustment system, wherein: the inherent mechanical stiffness is obtained through calibration using two steady-state loads, where F1 and F2 are the first and second steady-state loads, respectively, and z1 and z2 are the corresponding vertical displacements, and is based on (F 2- F1) / (z 2- z1) The inherent mechanical stiffness K of the seeding unit suspension system is calculated. sys .
[0012] The aforementioned adjustment system, wherein: the hydraulic cylinder applies an active thrust F to the seeding unit. hydro (k) Calculated based on the pressure difference between the two chambers, where A1 and A2 are the effective pressure-bearing areas of the two chambers, and P1(k) and P2(k) are the pressures of the two chambers.
[0013] The aforementioned adjustment system, wherein the comprehensive reaction force of the soil on the sown unit is decomposed into an equivalent stiffness term, an equivalent damping term, and a reference reaction force term, wherein Ks(k) is the soil equivalent stiffness, Cs(k) is the soil equivalent damping, and F0(k) is the soil reference reaction force.
[0014] The aforementioned regulation system, wherein the control device is configured to obtain soil reaction force observations by inverse calculation based on the machine-soil coupled dynamic equations and using the detected motion parameters and known active thrust.
[0015] The aforementioned adjustment system includes a control device used to predict the sowing depth fluctuation value by combining the soil equivalent parameters obtained from online identification of machine-soil coupled dynamics modeling with the current vibration state of the sowing unit.
[0016] The aforementioned regulation system, wherein the control device is further configured to perform soil dynamic parameter identification and to calculate the equivalent stiffness and equivalent damping of the soil in real time using a machine-soil coupled dynamic model.
[0017] The aforementioned adjustment system, wherein the real-time calculation of the equivalent stiffness and equivalent damping of the soil includes: Prediction: using parameters from the previous time step Multiply by the current state of motion Calculate an estimated reaction force; Comparison: Subtract the estimated reaction force from the actual reaction force y(k) to obtain the error e(k); Correction: Correct according to the error e(k) and the gain matrix. K s (k),C s (k),F 0 (k) ; Update: When When the force is insufficient to explain the total reaction force y(k), the remaining unexplained force is automatically attributed to F0(k).
[0018] The aforementioned adjustment system, wherein the prediction of seeding depth fluctuation values includes: Step A: Feature Vector Construction At each sampling time k, the controller calculates the total reaction force y(k) based on the motion parameters measured by the sensors and the hydraulic thrust, and then calls the current soil stiffness estimate and soil damping estimate. Construct the current 5-dimensional feature input vector ; Step B: Input via sliding window Set a historical sliding window of length N, and select N consecutive feature vector sequences from time k-N+1 to time k. As a whole tensor, it is input into the LSTM network; Step C: Nonlinear mapping and time series extrapolation The neural network calculates the predicted vertical displacement sequence of the seeded individual plants for the next T steps based on the changing trends of soil stiffness and damping, combined with the current pressing velocity, using a pre-trained weight matrix. , This represents the model's prediction of the vertical displacement of the seeding unit at the τth sampling time (τ = 1, 2, ..., T) under the information set conditions at the current sampling time k; T is the prediction time domain length.
[0019] The aforementioned adjustment system, wherein the controller adjusts the seeding unit in the following manner: 1) Prediction: Soil equivalent parameters obtained from online identification using machine-soil coupled dynamic modeling. The current vibration state of the seeding unit is input into the pre-trained time-series prediction model, and the vibration state is preferably selected. ,in For the vertical displacement of the seeding unit, Vertical velocity, The vertical vibration acceleration is calculated and compensated for by attitude. Predictive model outputs future Predicted values of seeding depth fluctuation within each sampling period:
[0020] in The reference displacement corresponding to the target seeding depth is represented by τ, where τ is the prediction step size index and k is the discrete sampling time index. 2) Indicator Calculation: The stability index is calculated using the following formula: in The upper limit of allowable seeding depth fluctuation, The upper limit of permissible vibration acceleration, Here are the normalized weighting coefficients, w1 is the displacement weight, and w2 is the acceleration weight; 3) Collaborative decision-making and execution: when At that time, it was assumed that the stability requirements were met, and the controller maintained the basic downward pressure of the downward-pressing cylinder. Unchanged; when At the same time, the controller calculates the virtual stiffness compensation component and the virtual damping vibration absorption component in parallel: For seeding depth deviation Calculate the virtual stiffness compensation force F1(k): Where Kg represents the virtual stiffness gain, This indicates the maximum prediction deviation of the sowing depth within the prediction window; For vertical vibration acceleration Calculate the virtual damping force F2(k): Where Cg represents the virtual damping gain, Finally, the controller generates the total execution instructions based on the stability index SI(k):
[0021] Where α1(k) and α2(k) ∈ [0,1] are the adjustment coefficients obtained by monotonic mapping of SI(k), and a synchronous adjustment strategy is adopted, i.e., α1(k) = α2(k) = α(k).
[0022] The aforementioned regulating system, wherein: The α(k) is calculated using a piecewise linear mapping function:
[0023] Where: SI0 is the initiation threshold for stability control; SI1 is the strong intervention threshold. When SI(k)≤SI0, the system is in the stable region, α(k) = 0, and the hydraulic system maintains a passive damping state without consuming additional power; when SI0 < SI(k) < SI1, the system is in the regulating region, and α(k) increases linearly or nonlinearly with SI(k), realizing variable gain flexible intervention; when SI(k)≥SI1, the system is in the instability risk region, α(k) = 1, and the actuator applies maximum stiffness and damping compensation.
[0024] The aforementioned adjustment system, wherein the controller adjusts the seeding unit, further includes: 4) Impact protection: When detected The system bypasses the aforementioned coordinated control, triggering rapid unloading / depressurization, a imp The impact trigger threshold; When detected Below the recovery threshold Continuous unloading holding time . Attached Figure Description
[0025] Figure 1 This is a schematic diagram of the overall hardware structure and installation location of the system of the present invention; Figure 2 Schematic diagram of the component structure of a parallel four-bar linkage; Figure 3 This is a schematic diagram of the dynamic model and force analysis of the seeder-soil coupling; Figure 4Flowchart of an active adjustment system for seeding depth stability based on a predictive model; Figure 5 Diagram illustrating how the prediction model works. Detailed Implementation
[0026] The following is in conjunction with the appendix Figure 1-5 The specific embodiments of the present invention will be described in detail below. These embodiments are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. Obviously, the embodiments described in this invention are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0027] The terms "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of the invention include the specific features, structures, or characteristics described in connection with that embodiment. Therefore, the terms "comprising," "including," "having," and variations thereof in this specification mean "including but not limited to," unless otherwise specifically emphasized.
[0028] like Figure 1 As shown, the precision seeder of this patent includes a frame 2, a seed box 1 mounted on top of the frame 2, and a press wheel 7, a soil covering wheel 8, a depth limiting wheel 10, a depth limiting wheel adjustment handle 3, a furrowing disc 11, and a stubble breaking wheel 12 mounted on the bottom of the frame 2. The stubble breaking wheel 12 is located at the head of the frame 2, the press wheel 7 is located at the tail of the frame 2, and the soil covering wheel 8 is located at the front of the press wheel 7. The depth limiting wheel 10 is mounted on the frame 2 via a depth limiting wheel arm 9. The depth limiting wheel adjustment handle 3 is mounted on the frame 2 and connected to the depth limiting wheel arm 9. The depth limiting wheel adjustment handle 3 is used to adjust the height difference between the depth limiting wheel 10 and the furrowing disc 11, i.e., to adjust the sowing depth. The press wheel 7 is connected to the frame 2 via a press cylinder 6, which provides pressing pressure to the press wheel 7.
[0029] The seed metering device 17 is mounted on the frame and located below the seed box 1. The pressure sensor 4 is installed at the connection between the depth limiting wheel arm adjusting handle 3 and the frame 2 to detect the pressure of the seed unit on the ground. The pressure sensor 5 is installed at the connection between the pressing cylinder 6 and the frame 2 to detect the supporting force of the ground on the pressing wheel 7.
[0030] A predictive model-based active adjustment system for seeding unit depth stability is installed on an independent seeding unit of a precision seeder. The seeding unit is mounted on the seeder's crossbeam via a parallel four-bar linkage 13, thus forming the seeder. The structure of the parallel four-bar linkage 13 is as follows: Figure 2 As shown.
[0031] The active adjustment system for seeding depth stability based on a predictive model includes a detection device, a control device, and an execution device. Among them: The detection device includes an IMU (Inertial Measurement Unit) 15, a pressure sensor 4, a pressure sensor 5, and an angle sensor 16.
[0032] The IMU (Inertial Measurement Unit) 15 is installed at the geometric center of the parallel four-bar linkage 13. It is a 6-axis MEMS sensor (MPU6050) with a sampling frequency of 200Hz, and is responsible for collecting the vertical acceleration and pitch angular velocity of the seeding unit.
[0033] The downward pressure sensor 4 is a pin-type load sensor, installed at the connection between the depth limiting wheel arm adjustment handle 3 and the frame 2, and is used to measure the supporting reaction force of the depth limiting wheel 10 on the ground.
[0034] The pressure sensor 5 is installed at the connection between the pressing cylinder 6 and the frame 2 to detect the supporting force of the ground on the pressing wheel.
[0035] Angle sensor 16 is installed at the hinge joint connecting the upper pull rod 18 of the parallel four-bar linkage 13 and the front fixed support of the frame 2. The detected value is used to calculate the relative height of the seed unit relative to the frame. Angle sensor 16 is used to monitor the rotation angle θ of the parallel four-bar linkage in real time. Since the link length of the parallel four-bar linkage is a known constant, the controller, based on the trigonometric geometric relationship (h = L sinθ), can accurately map the collected angle change to the vertical height change of the seed unit relative to the frame. This indirect measurement method effectively avoids the problem of non-contact distance sensors easily failing under dusty conditions.
[0036] The control device, installed in the cab of the traction machinery (such as a tractor), allows the operator to adjust and monitor the operating status of the seeding unit. The control device uses a high-performance embedded development board based on the ARM Cortex-M7 core and has a floating-point unit (FPU) for running complex dynamic models and neural network algorithms.
[0037] The actuator includes a pressing cylinder 6 and a lowering cylinder 14. The lowering cylinder 14 is mounted on the front rod 19 of the parallel four-bar linkage 13 at one end and on the lower rod 20 at the other end. The thrust is controlled by an electro-hydraulic proportional pressure reducing valve and is used to apply downward pressure to the seeding unit. A normally closed solenoid ball valve is connected in parallel in the hydraulic oil circuit as a quick unloading valve. The pressing cylinder 6 is mounted on the bracket of the pressing wheel 7 and is an electro-hydraulic cylinder used to apply dynamic pressing pressure to the pressing wheel 7.
[0038] The control device is used for machine-soil coupled dynamic modeling. like Figure 3As shown, to achieve stable control of sowing depth, the vertical motion of the sowing unit is equivalent to a single-degree-of-freedom system, and the machine-soil coupled dynamic equations are established: ; In the formula, M represents the equivalent mass of the seed unit (unit: kg). This represents the vertical vibration acceleration of the seeding unit, a value acquired by an inertial measurement unit (IMU) and obtained through attitude calculation and gravity compensation (unit: m / s²). For inertia; C sys K represents the inherent mechanical damping of the seeding unit suspension system (unit: N·s / m); sys Indicates the inherent mechanical stiffness of the seeding unit suspension system (such as a parallel four-bar linkage) (unit: N / m). These represent the inherent stiffness and damping of the structure, respectively, and are considered constants within the operational timescale, only to be recalibrated after "component wear / replacement / maintenance"; z(k) represents the vertical velocity of the seed unit (unit: m / s); z(k) represents the vertical displacement of the seed unit relative to its equilibrium position (unit: m); F soil (k) represents the combined reaction force of the soil on the seed unit (unit: N); F hydro (k) represents the active thrust (in N) applied by the hydraulic cylinder to the seeding unit; F down (k) represents the external downward pressure (generated by self-weight, counterweight / spring preload, or downward pressure device) (unit: N), and k represents the index of the discrete sampling time.
[0039] Calibration via off-ground steady-state loading: ; The damping ratio ζ was obtained through a free-attenuation test off the ground, and then... The calculation involves calculating the first and second steady-state loads applied to the seeding unit, F1 and F2 respectively, and the first and second vertical displacements of the seeding unit under these loads. The control device drives the hydraulic cylinder to apply a first-set steady-state thrust F1 to the seeding unit. After the system stabilizes, the first vertical displacement z1 is collected. Subsequently, the hydraulic cylinder is controlled to increase the driving force, applying a second-set steady-state thrust F2. After the system stabilizes, the second vertical displacement z2 is collected. Finally, the stiffness value is calculated based on Hooke's law. "System stability" refers to the complete decay and disappearance of the vertical vibration (transient response) excited by the seeding unit after being subjected to the step load applied by the hydraulic cylinder, with its velocity and acceleration returning to zero, ultimately resting in a static mechanical equilibrium state at a fixed displacement point.
[0040] The hydraulic thrust of the downward cylinder 14 is calculated from the pressure of the two chambers, using the following formula: A1 and A2 represent the effective pressure-bearing areas of the two chambers of the hydraulic cylinder (unit: m^2), and P1(k) and P2(k) represent the pressures of the two chambers of the hydraulic cylinder (unit: Pa or MPa).
[0041] To reflect the changes in reaction force caused by spatial variability of soil, an equivalent model of soil viscoelasticity is adopted:
[0042] Wherein, Ks(k) represents the equivalent soil stiffness (updated online as soil firmness / moisture content changes) (unit: N / m), and Cs(k) represents the equivalent soil damping (updated online as soil conditions change) (unit: N·s / m). As soil firmness and moisture content change, the sampling window can be updated with "segmented constant values".
[0043] F0(k) represents the soil baseline reaction force (also known as the baseline bias term). Physically, it characterizes the basic static support of the soil on the seeded unit (i.e., the resistance near zero displacement); algorithmically, it is a variable that updates slowly over time k. In the dynamic model, this invention absorbs the sensor's zero-point drift and low-frequency changes in soil firmness by updating F0(k) online.
[0044] By rearranging terms in the machine-soil coupled dynamic equations, soil reaction force observations are obtained, which are used to characterize the equivalent soil reaction force, and thus, real-time values of external excitations (data source) can be acquired. (1) After calculating the soil reaction force y(k) from the machine-soil coupled dynamic equation (Equation 1), in order to analyze the physical composition of this reaction force, an equivalent model of soil viscoelasticity is introduced (Equation 2): (2) The control device uses an online parameter identification algorithm to make the model estimate approximate the observed value y(k), thereby solving for the current soil stiffness in reverse. .
[0045] The update process (core of the RLS algorithm) is as follows: Within each sampling period Ts, preferably 1-20 ms, the control device performs a parameter update once: Prediction: Based on the model structure defined by formula (2), using the parameters from the previous time step. Multiply by the current state of motion Calculate the "theoretical estimated reaction force" at the current moment. θ specifically refers to the column vector consisting of the soil equivalent stiffness Ks(k), the soil equivalent damping Cs(k), and the soil reference reaction force F0(k), i.e. Correspondingly, θ(k-1) refers to the parameter estimate of the previous sampling period. The motion state vector corresponds one-to-one with the parameter θ, and its specific definition is: .in, The third element is always 1. This setting allows F0(k) to act as a 'bias term' in the algorithm, thus enabling it to continuously absorb static errors in the model.
[0046] At the algorithm startup time (k=0), the system has not yet come into contact with the soil or is in a standby state. At this time, the parameter vector is initialized to a zero vector, i.e. Specifically, at the algorithm startup time (k=0), F0(k) is initialized to 0. During subsequent operations, F0(k) iterates with sampling time k, dynamically increasing or decreasing based on the polarity and magnitude of the model's predicted residuals, thus adaptively approximating and representing the current soil baseline reaction force in real time. Then, during the iteration process, the algorithm utilizes... For terms where the constant value is 1, the DC component in the prediction error will be automatically accumulated and updated into F0(k), causing it to gradually converge from 0 to the current true soil reference reaction force.
[0047] Comparison: Compare the "actual observed reaction force" y(k) obtained by solving formula (1) with the "theoretical estimated reaction force" mentioned above. Subtraction yields the prediction error. .
[0048] Correction: Based on the prediction error e(k), the parameter vector is updated using the parameter update law of recursive least squares. Perform online correction. The calculation formula is as follows:
[0049] Where L(k) is the gain matrix calculated by the algorithm in real time. Using this formula, the algorithm dynamically allocates the model prediction error e(k) to the stiffness, damping, and reference reaction terms. Especially when the dynamic excitation (displacement and velocity changes) of the system is small, the error will primarily drive the... The update enables adaptive tracking of the static reference reaction force.
[0050] Stability prediction and control: like Figure 4 As shown, this embodiment proposes a closed-loop control method of "prediction-evaluation-coordinated control" for seeding depth stability. This method, executed by a control device, aims to reduce seeding depth fluctuations and minimize ineffective valve activation under conditions of spatial variability in soil firmness and ground disturbance. The control flow is as follows.
[0051] Specifically, the workflow of the prediction model includes the following steps: Step A: Feature Vector Construction At each sampling time k, the controller calculates the total reaction force y(k) (observed value) from the sensor measurements, and then calls the aforementioned dynamic identification module based on the RLS algorithm to obtain the current soil stiffness estimate and soil damping estimate. These two physical parameters serve as the input dimension of the neural network. Subsequently, they are compared with the motion state acquired in real-time by the sensor. Combine them to construct the current 5-dimensional feature input vector. .
[0052] Step B: Input via sliding window To incorporate historical trend information, the system maintains a historical sliding window of length N. This involves N consecutive feature vector sequences from time k-N+1 to time k. As a whole tensor, it is input into the input layer of the LSTM network.
[0053] Step C: Nonlinear mapping and time series extrapolation The LSTM network utilizes its internal memory units to calculate the predicted vertical displacement sequence of the seeding unit over future steps T, based on the input trends of soil stiffness and damping, combined with the current pressing velocity, and through a pre-trained weight matrix. . This represents the model's prediction of the vertical displacement of the seeding unit at the τth sampling time (τ = 1, 2, ..., T) under the information set conditions at the current sampling time k; T is the prediction time domain length.
[0054] The innovation of this approach lies in using soil parameters decoupled online from physical dynamics formulas as "prior knowledge" input to the network. Compared to end-to-end learning that directly uses raw sensor data, this scheme significantly reduces the network's dependence on the amount of training data and endows the neural network with explicit physical interpretability—that is, the model clearly understands whether the current instability is caused by "hard soil" or "high speed," thereby achieving more accurate predictions.
[0055] In the feature vector construction of step A, soil stiffness Ks (characterizing soil hardness) and compaction velocity (characterizing speed) are decoupled into two independent input dimensions [...,Ks(k), ..., [, ...] Parallel inputs are fed into the network. As Ks increases... When it remains unchanged, the network weights map it to soil quality changes; when When Ks increases while Ks remains constant, the network maps it to a perturbation in the operating speed, thus achieving accurate attribution and prediction at the physical mechanism level.
[0056] Furthermore, this invention also proposes a method for adjusting the seeding unit based on the predicted quantity.
[0057] 1) Prediction: Soil equivalent parameters obtained from online identification using machine-soil coupled dynamic modeling. (Read the soil equivalent stiffness and damping values that have been updated in step A and mark them as an estimated value sequence) and input the current vibration state of the sown unit into the pre-trained time series prediction model (e.g., Figure 5 In the example shown, the vibration state is preferred. .in The vertical displacement (or equivalent indentation) of the seeding unit. Vertical velocity, This represents the vertical vibration acceleration after attitude calculation and gravity compensation. The prediction model outputs the future... Predicted values of seeding depth fluctuation within each sampling period:
[0058] in The reference displacement (in mm) corresponding to the target depth is represented by τ, the prediction step size index (none), and k is the discrete sampling time index (none). Δz pred This indicates the relative z-displacement of the seeded unit within the prediction window. ref The maximum deviation (unit: mm), where, This represents the predicted vertical displacement of the seeding unit at time k, based on the time-series model output, for the future τ-th sampling time. ẑ represents the "predicted value" of the vertical displacement of the seeding unit, calculated by the LSTM network (neural network) algorithm in step C above. ref This is the system reference displacement obtained based on the preset target sowing depth calibration. Specifically, before operation, the user sets the target agronomic sowing depth (e.g., 50mm in this case) through the human-machine interface. The controller maps this depth value to the corresponding vertical displacement sensor value of the parallel four-bar linkage 13, based on the geometric parameters of the sowing unit and the sensor installation position, which is z. ref .
[0059] Compared with existing technologies: Traditional PID control can only handle e(k) = z(k) - z ref (That is: the current error).
[0060] The technical treatment of this invention is (i.e., future errors).
[0061] Conclusion: It is precisely because of this The existence of this predictor allows the system to act before errors occur, thus achieving "proactive adjustment".
[0062] 2) Index Calculation: To simultaneously reflect both "low-frequency large displacement caused by soft soil subsidence" and "high-frequency large impact caused by hard soil vibration," this invention constructs a stability index:
[0063] in The upper limit of allowable seeding depth fluctuation, The upper limit of permissible vibration acceleration, Here are the normalized weighting coefficients. w1 is the displacement weight, and w2 is the acceleration weight. In scenarios with soft soil or extremely high requirements for consistent sowing depth (such as precision corn sowing), it is preferable to increase the weight of the displacement prediction term, taking w1 = 0.6~0.7 and w2 = 0.3~0.4. This setting allows the system to focus on the trend z of sowing depth deviating from the target value. pred This is achieved by increasing virtual stiffness to maintain seeding depth stability.
[0064] 3) Collaborative decision-making and execution: When At that time, it was assumed that the stability requirements were met, and the controller maintained the basic downward pressure of the downward cylinder 14. Unchanged (F) base (k) represents the foundation pressure, the value of which is determined by the soil reference reaction force F0(k) obtained online, the seed unit weight M, and the gravitational acceleration g. base (k)=F0(k)-Mg, used to balance the static support of the soil); when At the same time, the controller calculates the virtual stiffness compensation component and the virtual damping vibration absorption component in parallel: Finally, the controller generates the total execution instruction F based on the stability index SI(k). cmd At the same time, the controller is also configured to perform pressurization coordination control: controlling the output pressure P of the pressurization cylinder 6. press Follow the output pressure P of the lower cylinder 14 down The changes, and their proportional relationships are as follows:
[0065] Wherein, λ is the compaction ratio coefficient. In this embodiment, considering both trenching resistance and the required soil compaction, the preferred method is... Adjustments should be made based on soil type: λ=0.3 in loose sandy soil to prevent excessive compaction; λ=0.5 in heavy clay or no-till soil to ensure effective closure of the planting furrow.
[0066] For seeding depth deviation Calculate the virtual stiffness compensation force F1(k):
[0067] Where Kg represents the virtual stiffness gain (unit: N / m). This indicates the maximum prediction deviation of the sowing depth within the prediction window. This indicator can reflect in advance the trend of individual sowing plants deviating from the target sowing depth in the future.
[0068] For vertical vibration acceleration Calculate the virtual damping force F2(k):
[0069] Where Cg represents the virtual damping gain (unit: N·s / m) Finally, the controller generates the total execution instructions based on the stability index SI(k):
[0070] Where Fcmd(k) represents the total hydraulic force command, the controller represents the total hydraulic force command, and the controller divides it by the effective area of the cylinder to convert it into a target pressure command before sending it to the hydraulic actuator. To ensure the stability of the closed-loop system, the virtual stiffness gain Kg and virtual damping gain Cg are selected according to the critical damping criterion: Selection of Kg: Set the natural frequency of the target system According to the formula The calculation yields the result. Typically, a weight (Kg) is chosen such that the closed-loop stiffness is slightly greater than the average equivalent stiffness of the soil, thus dominating the mechanical properties.
[0071] Selection of Cg: To prevent system overshoot, based on the critical damping formula... Determined. Where ξ is the target damping ratio, and in this embodiment, ξ = 0.707 is preferred. At this value, the system possesses optimal fast response and steady-state characteristics, and can dissipate the vibration energy input from hard soil impact most quickly. Where α1(k) and α2(k) ∈ [0,1] are adjustment coefficients obtained by monotonically mapping SI(k), α1(k) is the adjustment coefficient of the virtual stiffness compensation force F1(k), and α2(k) is the adjustment coefficient of the virtual damping vibration absorption force F2(k). In this embodiment, in order to ensure the synergistic effect of virtual stiffness and virtual damping, a synchronous adjustment strategy is adopted, that is, let α1(k) = α2(k) = α(k). This means that virtual stiffness and virtual damping are adjusted synchronously and proportionally, without decoupling calculation. The α(k) is preferably calculated using a piecewise linear mapping function:
[0072] This achieves the goal of "the more unstable the situation, the stronger the intervention." The force command can be converted into a target pressure difference or valve control current through the effective area of the hydraulic cylinder, and then output after amplitude limiting.
[0073] Using this formula, the system employs fine-tuning in the low-risk zone (SI slightly greater than SI0), meaning that when the SI index is low, the downward-pressing cylinder outputs only a small, mild compensating force (α is very small) to make slight attitude corrections to the seeding unit. In the high-risk zone (SI close to SI1), strong intervention is used, meaning that when the SI index is close to its limit, the downward-pressing cylinder outputs full-load stiffness and damping compensating force (α≈1), applying a strong downward pressure or vibration-absorbing force to the seeding unit to prevent it from bouncing. This variable gain strategy effectively avoids hydraulic shocks caused by abrupt changes in control commands, achieving a smooth transition in the control process.
[0074] Where: SI0 is the starting threshold for stability control, used to set the control dead zone to filter high-frequency noise and minor disturbances. In this embodiment, SI0 = 0.2 (i.e., allowing 20% normalized fluctuation margin); SI1 is the strong intervention threshold, used to define the full-load adjustment boundary. In this embodiment, SI1 = 1.0. When SI(k)≈0.2, the system is in the stable region, α(k) =0, and the hydraulic system maintains a passive damping state without consuming additional power; when 0.2 < SI(k) < 1.0, the system is in the adjustment region, and α(k) increases linearly or nonlinearly with SI(k), realizing variable gain flexible intervention; when SI(k) > 1.0, the system is in the instability risk region, α(k) = 1, and the actuator applies maximum stiffness and damping compensation.
[0075] 4) Impact protection: When an impact is detected... (a) imp Impact trigger threshold (unit: m / s^2), preferred When the system bypasses the coordinated control of step 3), when the pressure cylinder 14 is triggered to unload, due to the pressure following strategy, the pressure cylinder 6 also reduces its pressure synchronously, thereby assisting the entire unit in flexible obstacle avoidance; when Below the recovery threshold (a rec Recovery threshold (unit: m / s^2) duration (t hold After the unloading holding time (t), the above prediction and coordinated control are automatically restored: the entire closed-loop control process from step (1) to step (3) is restored. That is, after the impact ends (crossing the obstacle) and is held for a period of time (thold), the system exits the "soft float" mode and restarts: predicting the impact depth (step 1) → calculating the index (step 2) → applying the coordinated control force (step 3), that is, the system returns to the normal working state. Unloading holding time t hold The preferred value range is 0.1s to 0.5s, more preferably 0.2s.
[0076] Through the above method, the present invention utilizes online identification. As a predictive input, the controller can predict the instability trend of the seeding depth in advance under different soil conditions, and decompose the intervention into "equivalent stiffness type pressure compensation" and "equivalent damping type vibration absorption compensation", so as to realize adaptive and coordinated adjustment of two typical instability mechanisms: soft soil subsidence and hard soil vibration.
[0077] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A predictive model-based active adjustment system for seeding depth stability, used to control the seeding depth of a precision seeder, characterized in that: The system includes a detection device, a control device, and an execution device; the detection device is used to detect the motion parameters of the seeding unit; the control device is used to perform machine-soil coupling dynamic modeling based on the parameters detected by the detection device, predict the seeding depth of the seeding unit based on the modeling results, and control the execution device to work based on the predicted seeding depth; the execution device is used to drive the seeding unit under the control of the control device so that the seeding depth meets the predetermined depth.
2. The regulating system according to claim 1, characterized in that: The detection device includes an inertial measurement unit, a pressure sensor, a pressure sensor, and an angle sensor.
3. The regulating system according to claim 1, characterized in that: The actuator includes a pressing cylinder and a lowering cylinder.
4. The regulating system according to claim 1, characterized in that: The control device is used to equate the vertical motion of the seeding unit to a single-degree-of-freedom system and establish the machine-soil coupled dynamic equation.
5. The regulating system according to claim 4, characterized in that: The inherent mechanical stiffness is obtained through calibration with two steady-state loads, where F1 and F2 are the first and second steady-state loads, respectively, and z1 and z2 are the corresponding vertical displacements. The inherent mechanical stiffness K of the seeding unit suspension system is calculated based on (F2−F1) / (z2−z1). sys .
6. The regulating system according to claim 4, characterized in that: The hydraulic cylinder applies an active thrust F to the seeding unit. hydro (k) Calculated based on the pressure difference between the two chambers.
7. The regulating system according to claim 4, characterized in that: The combined reaction force of the soil on the seed unit is decomposed into an equivalent stiffness term, an equivalent damping term, and a reference term.
8. The regulating system according to claim 4, characterized in that: The control device is configured to obtain soil reaction force observations by inversely calculating based on the machine-soil coupled dynamic equations and using the detected motion parameters and known active thrust.
9. The regulating system according to claim 4, characterized in that: The control device is used to predict the seeding depth fluctuation value.