Electro-hydraulic control method and device for height of leaf vegetable harvester header

Abstract

The application provides a leaf vegetable harvester header height electro-hydraulic regulation method and device, and relates to the technical field of header height regulation. The application determines the allowable working pose space of the header within the working width as a safety boundary by constructing a predicted ground elevation profile in real time and setting a hierarchical control threshold, simultaneously evaluates a harvesting order index based on the torque and rotational speed signals of the driving motor of the conveying component, takes the index as an optimization target, and runs a multivariable closed-loop optimization controller under the constraint of the allowable pose space to automatically search for the optimal header pitch angle and ground clearance, and finally generates decoupling instructions to independently drive the pitch and lifting mechanisms through a decoupling controller based on an inverse dynamics model, thereby realizing real-time and accurate regulation of the header pose. The application combines ground surface topography prediction, harvesting state feedback and pose closed-loop optimization to realize adaptive optimization and regulation of the header height during the leaf vegetable harvesting process under complex terrain.

Description

Technical Field

[0001] This invention relates to the field of header height control technology, specifically to an electro-hydraulic control method and device for header height control of a leafy vegetable harvester. Background Technology

[0002] Mechanized harvesting of leafy vegetables is key to reducing production costs and improving industry efficiency. Unlike disorderly harvesting, orderly harvesting of leafy vegetables requires the harvester to precisely cut the roots while ensuring the vegetables enter the conveyor channel in a neat and controllable manner. This places extremely high demands on the harvester's operational posture control. However, leafy vegetable planting beds generally have complex terrain features such as unevenness and significant lateral elevation differences. During the harvester's movement, the cutting depth of the blades relative to the ground is constantly changing. Cutting too shallowly will result in missed cuts, while cutting too deeply will bring in a large amount of soil, causing blockages in the conveyor channel and even damaging the core of the vegetables, seriously affecting harvesting quality and operational efficiency.

[0003] Currently, automatic control methods for header height mainly rely on simple mapping between terrain detection and preset rules. For example, laser or ultrasonic sensors are used to perform single-line or sparse dot matrix scanning of the ground in front, and then the header's raising and lowering actions are triggered based on empirical thresholds. These methods have the following core limitations: First, they lack a complete perception of lateral ground elevation differences within the operating width, failing to address the problem of uncontrolled local cutting depth caused by lateral tilting of the ridge surface; second, the control strategies are mostly open-loop feedforward or simple threshold feedback, unable to proactively and continuously optimize the header's posture to adapt to dynamically changing ground and crop conditions; third, these methods completely ignore the working status of the internal conveying components of the header, failing to detect disordered states such as leafy vegetable accumulation, conveyor slippage, or foreign object impact caused by improper cutting depth, resulting in a broken information loop between header control and harvest quality. Often, a reactive response is only taken after blockages have occurred or the crop core has been damaged, failing to achieve preventative and precise control.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method and device for electro-hydraulic control of the header height of a leafy vegetable harvester, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for electro-hydraulic control of the header height of a leafy vegetable harvester, comprising the following steps: Step 1: Real-time acquisition of aircraft pitch angle, roll angle data and 3D point cloud of the ground surface of the area to be harvested; after spatiotemporal alignment and fusion, construct a predicted ground elevation profile with the current position of the cutter as the reference and including future time series. Step 2: Set layered control thresholds, including the minimum cutting depth to ensure the cutting of vegetable roots, the optimal cutting depth range to achieve orderly transport and minimal soil load, and the damage limit cutting depth to prevent damage to the vegetable heart; combine the layered control thresholds with the predicted ground elevation profile to determine the allowable working posture space of the cutting platform within the working width; the allowable working posture space defines the safety boundaries of the cutting platform's pitch angle and height above the ground. Step 3: During the harvesting operation, the torque and speed signals of the drive motor of the conveying component are collected in real time and analyzed in time and frequency. Characteristic parameters representing the harvesting state are extracted, and the harvesting orderliness index at the current moment is calculated through the preset harvesting orderliness evaluation model. Step 4: Using the harvest orderliness index as the optimization target and the allowable working pose space as the constraint, run a multivariable closed-loop optimization controller; the optimization controller adjusts the desired pitch angle and desired ground clearance of the header by continuously outputting perturbation signals, and automatically searches for the optimal header pose that keeps the harvest orderliness index at the target level based on the real-time response change trend of the harvest orderliness index. Step 5: Input the optimal cutter table pose into the preset decoupled controller based on the inverse dynamics model to generate mutually decoupled control commands, which independently drive the cutter table pitch adjustment mechanism and the lifting adjustment mechanism to achieve real-time control of the cutter table pose.

[0007] Furthermore, in the spatiotemporal alignment stage, the timestamp of each frame of the three-dimensional point cloud on the ground is paired with the pitch angle and roll angle of the machine in the same frame. Based on the traveling speed of the harvester, the point clouds at different times are transformed into the vehicle coordinate system with the current position of the cutter as the origin to obtain the pose-corrected point cloud data. The pose is defined by the pitch angle of the cutter and the height above the ground. The corrected point cloud data within the preset time window are stitched together and interpolated to form a predicted surface elevation curve covering the entire cutting platform operation width; this predicted surface elevation curve reflects the surface undulation and reflects the elevation difference between the rows to be harvested on the transverse profile.

[0008] Furthermore, the specific setting of the hierarchical control threshold is as follows: Based on the transverse profile of the predicted ground elevation surface at the current position of the cutter, and combined with the leafy vegetable variety, growth height and root depth, the minimum cutting depth value is determined; this minimum cutting depth value ensures that the actual cutting depth at any transverse position is greater than the preset allowable root residual length. Above the minimum cutting depth, based on the upper limit of the amount of soil attached to the roots and the requirement for stem integrity required for orderly conveying, the lower and upper limits of the optimal cutting depth range are determined so that when the cutter operates within this range, it not only cuts the vegetable roots but also ensures that the amount of soil attached does not exceed the preset threshold, while also ensuring that the clamping position of the conveying component on the leafy vegetables is in the middle of the stem. The damage limit cutting depth value is determined based on the height of the cabbage from the ground and the anti-blocking safety gap of the conveying channel; when the actual cutting depth reaches or exceeds the damage limit cutting depth value, it is determined that there is a risk of cutting the cabbage or blocking it.

[0009] Furthermore, the construction of the permitted work pose space specifically involves: On the predicted ground elevation surface, all combinations of cutter pitch angles and ground clearance are traversed within the working width; combinations that would cause the predicted cut depth at any lateral position to exceed the damage limit cut depth value or fall below the minimum cut depth value are eliminated; the remaining combinations are defined as the allowed working pose space.

[0010] Furthermore, the construction and calculation of the harvest orderliness assessment model specifically includes: The characteristic parameters include: the energy ratio of the drive motor torque signal within a preset frequency band, the low-frequency fluctuation amplitude of the speed signal, and the coherence coefficient between the torque and speed signals; The harvest orderliness evaluation model is a weighted fusion function or a pre-trained machine learning model. Its input is the feature parameters calculated in real time, and its output is a harvest orderliness index between 0 and 1. Here, 1 represents an ideal ordered state, and 0 represents a completely blocked or severely abnormal state.

[0011] Furthermore, the workflow of the multivariable closed-loop optimization controller is as follows: Set a target threshold or target range for the harvest orderliness index, and superimpose a sine wave or pseudo-random wave of preset amplitude as a tentative pose command based on the expected pitch angle and expected ground height of the current header. Execute the tentative pose command, and perform correlation analysis between the superimposed sine wave or pseudo-random wave in the tentative pose command and the real-time response of the monitored harvest orderliness index; determine the gradient direction of the harvest orderliness index with respect to the header pitch angle and ground clearance by comparing the phase relationship between the two, and determine the gradient magnitude by the ratio of the response amplitude to the amplitude of the superimposed sine wave or pseudo-random wave, thereby forming a complete performance index gradient vector; Along the gradient vector direction, with an adaptive step size positively correlated with the gradient magnitude, update the desired pitch angle and desired ground clearance to generate a new cutter pose command; the new cutter pose command is the optimal cutter pose obtained in this optimization. Before outputting the new cutting table pose command, verify whether it falls within the allowed working pose space. If it exceeds the safety boundary, project it to the nearest point on the boundary of the allowed working pose space. The above process is continuously executed in a loop to form a closed-loop optimization control with the harvest orderliness index as the optimization objective.

[0012] Furthermore, when the multivariate closed-loop optimization controller fails to make the harvest orderliness index reach the target threshold or target range within a preset time period, the system triggers a graded response mechanism; the mechanism includes: issuing a warning to the operator, reducing the harvester's travel speed, or performing a safe shutdown.

[0013] Furthermore, the decoupling controller embeds an inverse dynamics model that characterizes the dynamic coupling relationship between the pitch and rise motions of the cutter table. This inverse dynamics model takes the optimal cutter table pose as input and calculates a feedforward control command to compensate for the cross-coupling effect of the two sets of hydraulic actuators, so as to drive the actual pose of the cutter table to track the optimal cutter table pose within a preset overshoot allowable range.

[0014] Furthermore, the specific construction and solution method of the inverse dynamics model is as follows: A positive dynamic model of the electro-hydraulic servo system of the cutting platform is established. This positive dynamic model uses the pitch angle and ground clearance of the cutting platform as generalized coordinates. It describes the dynamic behavior of the cutting platform under the simultaneous drive of the pitch adjustment mechanism and the height adjustment mechanism through the inertia matrix, the Coriolis force and centrifugal force matrix, and the gravity vector. The off-diagonal elements of the inertia matrix and the Coriolis force and centrifugal force matrix represent the cross-coupling effect of the motion of the pitch adjustment mechanism and the height adjustment mechanism. The input of the positive dynamic model is the driving force and the driving torque, and the output is the acceleration of the ground clearance and pitch angle of the cutting platform. The positive dynamics model is mathematically inverted to obtain the inverse dynamics model; the input of the inverse dynamics model is the desired cutter pose and the velocity and acceleration of the cutter pose, and the output is the desired driving force and desired driving torque required to achieve the desired cutter pose. The optimal cutter pose, along with the velocity and acceleration calculated from the optimal cutter pose through differential or filtering, are substituted into the inverse dynamics model to calculate and output the corresponding desired driving force and desired driving torque. Based on the pre-calibrated hydraulic cylinder force-pressure mapping relationship and the flow-pressure-current characteristics of the electro-hydraulic servo valve, the desired driving force and desired driving torque are sequentially converted into the target driving pressure of each hydraulic cylinder, and then converted into the feedforward control current signal to drive the servo valve. This feedforward control current signal is the feedforward control command generated by the decoupling controller.

[0015] The present invention also provides an electro-hydraulic control device for the height of the header of a leafy vegetable harvester, which is used to implement the above-mentioned electro-hydraulic control method for the height of the header of a leafy vegetable harvester, comprising: The surface topography prediction module is used to collect real-time data on the aircraft's pitch angle, side tilt angle, and the three-dimensional point cloud of the surface of the area to be harvested. After spatiotemporal alignment and fusion, it constructs a predicted surface elevation profile based on the current position of the cutter and including future time series. The safety boundary construction module is used to set layered control thresholds, including the minimum cutting depth to ensure the cutting of vegetable roots, the optimal cutting depth range to achieve orderly transportation and minimal soil carrying, and the damage limit cutting depth to prevent damage to the vegetable heart; the layered control thresholds are combined with the predicted ground elevation profile to determine the allowable operating posture space of the cutting platform within the operating width; the allowable operating posture space defines the safety boundaries of the cutting platform's pitch angle and height above the ground; The harvest orderliness assessment module is used to collect torque and speed signals of the drive motor of the conveying component in real time during the harvesting operation, perform time-frequency analysis, extract characteristic parameters that characterize the harvesting state, and calculate the harvest orderliness index at the current moment through a preset harvest orderliness assessment model. The pose closed-loop optimization module is used to run a multivariable closed-loop optimization controller with the harvest orderliness index as the optimization target and the allowable working pose space as the constraint. The optimization controller adjusts the desired pitch angle and desired ground clearance of the header by continuously outputting perturbation signals, and automatically searches for the optimal header pose that keeps the harvest orderliness index at the target level based on the real-time response change trend of the harvest orderliness index. The decoupled execution control module is used to input the optimal cutter table posture into a preset decoupled controller based on an inverse dynamics model, generate mutually decoupled control commands, and independently drive the cutter table pitch adjustment mechanism and the lifting adjustment mechanism to achieve real-time control of the cutter table posture.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention constructs a predicted surface elevation curve containing future time series, fully perceiving the differences in lateral and longitudinal surface undulations within the working width, effectively solving the problem that traditional single-line scanning cannot cope with the lateral tilt of the ridge surface. Based on this, by setting three control thresholds—minimum cutting depth, optimal cutting depth range, and damage limit cutting depth—and constructing a permissible working posture space as a safety boundary, it ensures that the cutter will not break through the bottom line of damaging the heart of the vegetable or missing the roots under any terrain conditions. This invention also incorporates the internal conveying state of the header into the control closed loop by analyzing the torque and speed signals of the drive motor of the conveying component in real time, establishing a feedback link from header posture to harvest quality. The multivariable closed-loop optimization controller, with the harvest orderliness index as the optimization objective, can actively and continuously search within the permissible posture space for the optimal header pitch angle and ground clearance that makes the conveying smoothest and most orderly, effectively preventing disordered states such as root accumulation and conveyor belt slippage. With the help of a decoupled controller based on an inverse dynamics model, this invention achieves precise pose tracking while eliminating the cross interference between pitch and rise motions, making the cutting platform move faster and more smoothly, and ultimately realizing the orderly, low-loss and efficient harvesting of leafy vegetables in complex terrain. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall method flow of the present invention; Figure 2 This is a schematic diagram showing the changes in the harvesting orderliness index and ground clearance during harvesting operations in this invention; Figure 3 This is a schematic diagram showing the relationship between the harvest orderliness index and the torque of the drive motor during the harvesting operation in this invention; Figure 4 This is a schematic diagram of the device structure of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] Example: Please see Figures 1 to 3 The present invention provides a technical solution: A method for electro-hydraulic control of the header height of a leafy vegetable harvester, comprising the following steps: Step 1: Real-time acquisition of aircraft pitch angle, roll angle data and 3D point cloud of the ground surface of the area to be harvested. After spatiotemporal alignment and fusion, a predicted ground elevation profile is constructed based on the current position of the cutter and includes future time series.

[0021] At the front end of the harvester, a 3D vision sensor (such as a binocular stereo camera, LiDAR, or TOF camera) and an inertial measurement unit (IMU) are fixedly installed. The 3D vision sensor acquires three-dimensional point cloud data of the ground surface of the area to be harvested at a fixed frame rate (e.g., 30 Hz), and its field of view covers the entire working width of the cutter head and a distance in front of it (e.g., 0.5 meters to 1.5 meters). The inertial measurement unit measures and outputs the pitch and roll angle data of the harvester in real time.

[0022] The 3D vision sensor and the inertial measurement unit are connected via a hardware synchronization trigger line or by receiving a clock signal from the same external clock source, ensuring that each frame of point cloud data and each frame of inertial measurement data is given a unified timestamp with nanosecond-level precision.

[0023] Perform the following operations on the control unit: First, the timestamp of each frame of the 3D point cloud on the ground is paired with the nearest inertial measurement unit (IMU) data frame (containing the aircraft's pitch and roll angles). Specifically, using the point cloud's timestamp as a reference, the nearest IMU data frame is searched backward. If the time difference between the two is less than a preset time alignment tolerance, for example, less than 5 milliseconds, the pairing is considered successful; otherwise, the point cloud frame is interpolated or discarded. This pairing operation establishes the precise attitude of the sensor relative to the horizontal plane during the acquisition of each frame of point cloud.

[0024] Next, based on the harvester's current speed, the distance the harvester has traveled from the point cloud acquisition time to the current calculation time is calculated. Using this distance, the point clouds acquired at different times are translated and transformed from their respective sensor historical coordinate systems to a vehicle coordinate system with the current position of the cutter as the origin. This vehicle coordinate system is defined as follows: the origin is the projection of the lowest point of the cutter onto the ground; the positive X-axis aligns with the harvester's direction of travel; the positive Y-axis points to the left of the harvester's direction of travel according to the right-hand rule; and the positive Z-axis points vertically upwards. This coordinate system definition is naturally consistent with the harvester's own pose perception and adjustment direction, simplifying subsequent calculations.

[0025] Finally, by utilizing the paired fuselage pitch and roll angles, the point cloud, which is essentially tilted due to the vehicle's attitude, is restored to a vertical view with the horizontal plane as the reference, thus obtaining a corrected point cloud that reflects the actual terrain undulations. The specific process is as follows: Define the vehicle coordinate system and attitude angles: With the harvester in a horizontal, stationary state as a reference, define the vehicle coordinate system so that the X-axis points in the direction of vehicle movement, the Y-axis points to the left, and the Z-axis points upwards. The fuselage pitch angle is defined as the angle of rotation of the vehicle body around the Y-axis. The positive angle is defined as the angle at which the front of the vehicle is tilted upwards; the side tilt angle is defined as the angle at which the vehicle body rotates around the X-axis. The angles are taken as the rightward tilt of the vehicle body (i.e., the left side is raised), and these two angles are provided in real time by the inertial measurement unit.

[0026] Construct the total rotation matrix The rotation order of Euler angles ZYX is used here, but in this embodiment, the yaw angle (rotation about the Z-axis) does not participate in attitude compensation, and the roll angle... For small angles, pitch angle This is the primary attitude change variable. Therefore, the total rotation matrix... It is formed by multiplying the basic rotation matrices about the X-axis (tilt) and about the Y-axis (pitch) in sequence: Roll angle around the X-axis rotation matrix for: Rotate pitch angle around Y-axis rotation matrix for: Total rotation matrix: The total rotation matrix Multiplying a three-dimensional coordinate point vector by its left side physically means: first rotate that point around the X-axis. Angle, then rotate the result around the Y-axis. Angle. Since the compensation is for the vehicle's attitude, the point cloud collected by the sensors, which follows the vehicle's tilt, needs to be rotated back to a horizontal state.

[0027] Finally, reverse attitude compensation is performed. This is done for any point in the point cloud that has been translated to the current cutter coordinate system. Its coordinates are Perform the following calculations: in, Point coordinates The coordinate vector of the point after pose compensation is called the coordinate vector. Let be the inverse of the total rotation matrix R. Since the rotation matrix is ​​orthogonal, its inverse is equal to its transpose, i.e. Therefore, it can be used directly in actual calculations. Transformation is performed to save computing resources.

[0028] After reverse attitude compensation, the pose-corrected point cloud data is obtained.

[0029] In this embodiment, the position of the cutting platform is a set of two independent and orthogonal degree-of-freedom parameters, specifically the pitch angle of the cutting platform relative to the current horizontal ground and the height of the cutting tool above the ground. Their physical meanings are as follows: The pitch angle of the cutting platform is defined as the angle between the cutting platform's own reference plane (such as the bottom surface of the cutting platform or the plane where the blade beam is located) and the current horizontal ground. It is agreed that the direction of the cutting platform's front end tilting downward, i.e. pointing to the bottom, is negative, and the direction of the cutting platform's front end tilting upward is positive. This angle directly determines the angle at which the cutter enters the soil and the degree of inclination of the conveying trough.

[0030] The ground clearance is defined as the distance from the lowest point of the cutter along the vertical direction to the current horizontal ground, and it is agreed that a value above the ground is positive. This distance directly determines the absolute depth of the root cutting operation.

[0031] Before stitching the point cloud, preprocessing of the pose-corrected point cloud is also included. Specifically, a statistical outlier removal algorithm is used to statistically analyze the average distance within the neighborhood of each point, removing sparse noise points whose mean distance exceeds twice the standard deviation. These noise points are usually caused by dust, splashing water droplets, or crop leaves. At the same time, a voxel grid downsampling method is used, such as dividing the space into a cubic grid with a side length of 1 cm, and using the centroid of all points in each grid to represent the points of that grid, converting the unstructured point cloud into a structured raster map, thereby reducing the computational load and ensuring the smoothness of elevation changes.

[0032] Preprocessed corrected point cloud data within a preset time window (e.g., the past 0.5 seconds) are stitched together and spatially interpolated to form a predicted surface elevation covering the entire cutting platform operation area. The spatial interpolation preferably uses inverse distance weighted interpolation, and its calculation formula is as follows: Where Z is the predicted elevation value of the grid point to be interpolated; The elevation value of the i-th valid point cloud point that falls within the search radius (e.g., 10 cm) of this grid point; The horizontal distance from the i-th point cloud point to the raster point to be interpolated; is the power of the distance; i is the index of the valid point cloud point. The effective number of point cloud points; wherein, the effective point cloud points refer to point cloud points that fall within the search radius of the grid points to be interpolated.

[0033] In this example, the power exponent The value is 2. In agricultural settings, terrain undulations are typically continuous and smooth. The resulting inverse quadratic proportionality assigns significantly higher weights to nearby points, ensuring accurate restoration of terrain details, while avoiding excessive interference from distant points. This results in smooth, abrupt interpolation results that accurately reflect the physical reality of farmland surface morphology. When the number of point cloud points within the search radius is less than a preset threshold (e.g., 3), the grid point is retained as unobserved and not interpolated, thus preventing the generation of unreliable inference data.

[0034] This predicted surface elevation surface reflects the undulations of the land and shows the elevation differences between rows to be harvested in the transverse profile, such as the undulations of the ridge surface, furrows, and crop rows. This surface is a predicted surface elevation profile based on the current position of the cutter and containing future time series, informing the control system in advance how the terrain ahead will change, and providing a data basis for the advance and smooth adjustment of the cutter head posture.

[0035] During the stitching process, a Kalman filter is further used to perform time-series filtering on the elevation of each grid. The difference elevation calculated in each frame is used as the observation value to update the historical state, thereby outputting a dynamic elevation map that is smoother, more stable and predictive in time series.

[0036] Step 2: Set layered control thresholds, including the minimum cutting depth to ensure the cutting of vegetable roots, the optimal cutting depth range to achieve orderly transport and minimal soil carrying, and the damage limit cutting depth to prevent damage to the vegetable heart; combine the layered control thresholds with the predicted ground elevation profile to determine the allowable operating posture space of the cutting platform within the operating width; the allowable operating posture space defines the safety boundaries of the cutting platform's pitch angle and height above the ground.

[0037] In this embodiment, the setting of the hierarchical control threshold is as follows: Get the root depth of the currently harvested leafy vegetable variety ,Should The average vertical distribution depth of the root system of this variety in the soil at maturity, obtained in advance through field sampling, is expressed in millimeters.

[0038] Set the allowable length of root remnants This refers to the maximum length of vegetable roots that are allowed to remain in the soil after harvesting, measured in millimeters. The value is determined according to agronomic requirements, for example, it is set to 10 mm to 20 mm, and in this embodiment it is 15 mm.

[0039] The minimum cutting depth is calculated using the following formula: in, Minimum cutting depth value, representing the theoretical minimum depth to which the cutting tip penetrates the soil from the ground surface, in millimeters.

[0040] For example, if a certain leafy vegetable variety It is 80 mm. If it is 15 millimeters, then Millimeters. The rationale for this value is that it is only appropriate when the actual cutting depth of the cutter reaches or exceeds [the specified value]. Only then can the main body of the root system be cut off, ensuring that the length of the remaining roots underground does not exceed [a certain value]. This ensures that the plants can be properly tilled or prepared for subsequent land use, while also preventing them from falling over or being missed due to shallow cutting.

[0041] Above the minimum cutting depth, based on the upper limit of the root soil-bearing volume required for orderly transport and the stem integrity requirements, the lower and upper limits of the optimal cutting depth range are determined. The specific process for determining the optimal cutting depth range is as follows: Define soil carrying capacity threshold This represents the upper limit of the soil mass attached to the roots of a single leafy vegetable plant, expressed in grams. This threshold is determined based on the tolerance of subsequent washing processes or commercial requirements. In this embodiment... Take 30 grams.

[0042] Define a stem clamping safety zone, which is a preset length range extending upwards from the base of the stem, based on the total length of the stem. The percentage indicates that, in this embodiment, the safe zone is set to 20% to 60% of the total stem length. That is, the clamping position should fall within this range to avoid clamping too low and damaging the base or clamping too high and causing slippage during transport.

[0043] The relationship curves between cutting depth and soil amount and between cutting depth and clamping position were established through field pre-experiments. The field pre-experiments were conducted under the standard planting conditions of this leafy vegetable variety, and harvesting experiments were carried out at different cutting depths. The average soil amount per plant at each cutting depth and the stem height corresponding to the actual clamping position of the conveying component after the cutter cuts the vegetable root were measured. After repeated measurements, the average values ​​were taken to obtain two sets of discrete data points. Then, a continuous relationship curve was obtained by polynomial fitting.

[0044] The lower limit of the optimal cutting depth interval is determined based on the two relationship curves mentioned above. and upper limit : Take the minimum cutting depth that simultaneously satisfies the following two conditions: the amount of soil carried does not exceed The clamping position enters the lower boundary of the safe clamping zone of the stem, that is, at 20% of the total length of the stem.

[0045] Take the maximum cutting depth that simultaneously satisfies the following two conditions: the amount of soil carried does not exceed The clamping position should not exceed the upper boundary of the safe clamping zone of the stem, that is, 60% of the total length of the stem.

[0046] For example, if, through testing, a certain variety of leafy vegetable is found to have a soil-bearing weight of 28 grams at a cutting depth of 60 mm and a clamping position at 25% of the stem, a soil-bearing weight of 32 grams at a cutting depth of 90 mm and a clamping position at 55% of the stem, and a soil-bearing weight of 38 grams at a cutting depth of 100 mm and a clamping position at 65% of the stem, then the optimal cutting depth range is determined to be [60, 85] mm because the soil-bearing weight at 90 mm is close to the threshold, and both the soil-bearing weight and clamping position at 100 mm exceed the standard. The rationale behind this method is that it replaces empirical judgment with quantifiable physical quantities (soil-bearing weight, clamping position percentage), making the determination of the optimal cutting depth range reproducible and adaptable to different varieties.

[0047] The damage limit cutting depth value is determined based on the height of the Chinese cabbage above the ground and the anti-blocking safety clearance of the conveying channel. The specific process is as follows: Get the height of the Chinese cabbage from the ground ,Should This refers to the average vertical height of the heart of this leafy vegetable variety from the ground surface during its optimal harvest period, measured in millimeters, obtained through field measurements.

[0048] Set anti-blocking safety gap ,Should The minimum clearance height required at the entrance of the conveying channel to avoid blockage by soil and stubble accumulation is measured in millimeters and is determined based on the structural dimensions of the conveying channel and soil flowability tests. In this embodiment, it is taken as 30 millimeters.

[0049] Then the damage limit cutting depth value Calculate using the following formula: in, The damage limit cutting depth value represents the maximum permissible depth to which the cutting tip can cut downwards from the ground surface, expressed in millimeters.

[0050] For example, if The measured value was 50 millimeters. Take 30 mm, then =50-30=20 mm. The rationale for this value is that when the actual cutting depth of the cutter reaches... At that time, the tip of the knife was only a few inches away from the growing point of the cabbage. The safe distance; if the cutting depth is further increased, there is a risk of the cutter directly damaging the heart of the vegetable or the pile height of the chopped soil and stubble exceeding the safe distance. This leads to a dual risk of blockage at the entrance of the conveyor channel. In actual operation, the cutting depth of the cutter is positive when it is below the ground surface, and negative when it is above the ground surface; in this embodiment... A small positive value, or possibly a negative value depending on the variety, indicates that the cutter cannot penetrate too deeply into the soil, and may even need to cut above the surface.

[0051] When the actual cutting depth reaches or exceeds the damage limit cutting depth value, it is determined that there is a risk of cutting the heart of the vegetable or causing blockage.

[0052] In this embodiment, the construction of the allowed working pose space specifically involves: On the predicted ground elevation surface, all combinations of the pitch angle and ground clearance of the cutting platform within the operating width are traversed, specifically: Define the adjustable range of the header's pitch angle and ground clearance: The adjustable range of the header's pitch angle is determined by the design stroke of the header's pitch adjustment mechanism, typically from -15° to +15°, where negative values ​​indicate the header's front end is tilted downwards (pitch angle), and positive values ​​indicate the header's front end is tilted upwards (elevation angle). The adjustable range of the ground clearance is determined by the design stroke of the header's lifting adjustment mechanism, typically from -100 mm to +200 mm, with the cutter's position on the ground as the zero point, and upwards as positive.

[0053] Discretize the two continuous ranges mentioned above to generate a set of pose combinations to be traversed. The step size is set according to the control accuracy requirements, with a pitch angle step size of 0.5° and a ground clearance step size of 5 mm.

[0054] For each pose combination, the cutting depth of the cutter relative to the predicted ground elevation surface at each lateral position along the working width is calculated at that pitch angle and ground height. The cutting depth is defined as the elevation of the cutter tip minus the elevation of the predicted ground elevation surface at the corresponding lateral position; a positive value indicates that the cutter tip is in the soil, and a negative value indicates that the cutter tip is above the ground surface.

[0055] Iterate through all pose combinations and determine for each combination: if the cut depth at any lateral position is greater than the damage limit cut depth value. Or, the depth of cut at any lateral position is less than the minimum cutting depth value. If a position combination is deemed infeasible, it is marked as infeasible; otherwise, it is marked as feasible. The set of all position combinations marked as feasible constitutes the permitted operational position space. This permitted operational position space essentially defines the permissible posture boundaries of the header while maintaining safe and effective harvesting.

[0056] In this embodiment, the working pose space is allowed to be updated in real time as the harvester moves. Whenever the predicted ground elevation surface in step 1 is updated, the cut-in depth at each lateral position is recalculated based on the new ground elevation profile, and the allowable working pose space is regenerated to ensure the timeliness of the safety boundary.

[0057] Step 3: During the harvesting operation, the torque and speed signals of the drive motor of the conveying component are collected in real time and analyzed in time and frequency. Characteristic parameters representing the harvesting state are extracted, and the harvesting orderliness index at the current moment is calculated through the preset harvesting orderliness evaluation model.

[0058] The conveying components include a clamping conveyor belt, a reel, or a screw conveyor. The drive motor is preferably a permanent magnet synchronous motor or a brushless DC motor. Its torque signal is obtained through vector transformation calculation using a current sensor built into the motor driver, and the speed signal is acquired through an incremental encoder or Hall sensor mounted at the motor tail. The sampling frequency for both torque and speed signals is set to 500 Hz to ensure that transient changes caused by fluctuations in crop feed rate can be captured.

[0059] In this embodiment, the construction and calculation of the harvest orderliness assessment model specifically includes: The characteristic parameters include: the energy proportion of the drive motor torque signal within a preset frequency band, the low-frequency fluctuation amplitude of the speed signal, and the coherence coefficient between the torque and speed signals. The extraction methods for these three characteristic parameters are as follows: Feature parameter 1: The energy percentage of the torque signal within the preset frequency band; A Fast Fourier Transform (FFT) was performed on the real-time torque signal of the drive motor to obtain the power spectral density of the torque signal. The sampling time of the torque signal was 1024 of the most recent sampling points for each analysis, i.e., a data window of about 2 seconds. The Hanning window function was selected to reduce spectral leakage.

[0060] A preset frequency band is defined. Field trials have shown that when leafy vegetable transport is orderly, torque fluctuations are mainly concentrated in the low-frequency range of 0 to 5 Hz, corresponding to the periodic cutting force of the cutter and the stable load changes of the conveyor belt. However, when abnormalities such as blockages, unstable clamping, or excessive soil load occur, significant energy spikes in the torque signal appear in the 5 to 30 Hz frequency band. Therefore, this embodiment sets the preset frequency band to 5 to 30 Hz.

[0061] The energy percentage of the torque signal within the preset frequency band is calculated using the following formula and denoted as: : in, The power spectral density of the torque signal at frequency f is the sum of the power spectral densities at each frequency point within the preset frequency band of 5 to 30 Hz, and the denominator is the sum of the power spectral densities of the entire frequency band within the Nyquist frequency range of 250 Hz. Dimensionless values ​​between 0 and 1; The range of summation is , The range of summation is .

[0062] Energy percentage The meaning of this value is: the more orderly the conveying process and the more uniform the feeding, the smaller the proportion of mid-to-high frequency fluctuation energy in the torque signal. The closer it is to 0, the higher the proportion of medium- and high-frequency energy will be when abnormalities such as intermittent blockages, foreign object entry, or conveyor belt slippage occur. The corresponding increase.

[0063] Feature parameter 2: Low-frequency fluctuation amplitude of the speed signal; The real-time speed signal is detrended; the DC component and slowly changing trend term in the speed signal are removed by first-order difference method or high-pass filtering, and the cutoff frequency is set to 0.5 Hz, thereby separating the speed fluctuation component.

[0064] Bandpass filtering was performed on the fluctuation component in the 0.5 to 5 Hz frequency band to extract the speed fluctuation amplitude sequence in this frequency band; a second-order Butterworth filter was selected as the bandpass filter.

[0065] Calculate the root mean square value of the speed fluctuation amplitude within the preset observation window, which is taken as the low-frequency fluctuation amplitude of the speed signal, and denoted as . The unit is revolutions per minute (rpm). The observation window length is consistent with the torque signal analysis window, which is 1024 sampling points. The calculation formula is: Where i = 1, 2, ..., N, i is the index of the sampling point in the observation window, N is the number of sampling points in the observation window, and N = 1024; This represents the i-th speed fluctuation value after detrending and bandpass filtering.

[0066] Low-frequency fluctuation amplitude of speed signal The meaning of the value is as follows: when the clamping and conveying of leafy vegetables by the conveying component is in an orderly state, the load changes smoothly and the speed fluctuation is small; when the clamping position is improper, causing the plant to slip or get stuck, the drive motor will experience a momentary drop in speed or overshoot due to sudden load changes, and the low-frequency fluctuation amplitude will increase significantly.

[0067] Characteristic parameter 3: Coherence coefficient between torque and speed signals; The cross-power spectral density and individual self-power spectral density of the torque and speed signals were calculated in the 0 to 250 Hz frequency band. The same 1024-point data window and Hanning window as described above were used, and the Welch average periodogram method was employed, with 4 segments and 50% overlap between each segment to reduce the estimation variance.

[0068] The coherence coefficient between the torque and speed signals is calculated using the following formula, denoted as: Its value is also a dimensionless value between 0 and 1: in, This is the peak frequency point where the torque signal and the speed signal have the highest coherence in the 0 to 250 Hz frequency band. This represents the cross-power spectral density of torque and rotational speed at that frequency point. and These are the self-power spectral densities of torque and rotational speed at that frequency point, respectively.

[0069] Coherence coefficient The meaning of the value is: when the conveying process is normal and orderly, the torque and speed mainly exhibit a deterministic electromechanical coupling relationship determined by the load characteristics, and the two have high coherence at a certain characteristic frequency. Approaching 1; when abnormal operating conditions occur, such as partial blockage of the conveying channel or excessive soil carrying leading to a sudden increase in friction, the system's nonlinearity increases, and the linear correlation between torque and speed decreases. Significant decrease.

[0070] The harvest orderliness evaluation model is a weighted fusion function or a pre-trained machine learning model. Its input is the feature parameters calculated in real time, and its output is a harvest orderliness index between 0 and 1. Here, 1 represents an ideal ordered state, and 0 represents a completely blocked or severely abnormal state.

[0071] This embodiment uses a weighted fusion function as the specific implementation of the harvest orderliness evaluation model, as detailed below: The three feature parameters mentioned above are normalized to map them uniformly to the dimensionless interval of 0 to 1, with consistent direction. That is, the closer the parameter value is to 1, the more ordered the state; the closer it is to 0, the more abnormal the state. The normalization method is as follows: Regarding the energy ratio of the torque signal Since smaller values ​​indicate greater order, reverse normalization is used: in, This represents the normalized value indicating the proportion of torque signal energy. The preset upper limit threshold for energy percentage is based on multiple measurements and statistics under unblocked no-load conditions. The maximum value is set to 0.3 in this embodiment. This upper limit threshold is determined based on the following: under stable no-load operation, the proportion of medium- and high-frequency energy typically does not exceed 0.1%; considering that crop feeding under normal harvesting conditions will cause a certain degree of torque fluctuation, a certain margin is reserved before... Set to 0.3; when When this value is exceeded, the proportion of mid-to-high frequency energy is considered to have significantly deviated from the normal range. Take 0.

[0072] For the low-frequency fluctuation amplitude of the speed signal Since smaller values ​​indicate greater order, reverse normalization is used: in, The normalized value representing the low-frequency fluctuation amplitude of the speed signal; The preset upper limit threshold for speed fluctuation is based on the critical point of blockage observed in multiple blockage tests. The value is determined; in this embodiment, it is set to 150 revolutions per minute. The basis for determining this upper limit threshold is that, under normal conveying conditions, the rotational speed fluctuation range is typically maintained within the range of 30 to 60 revolutions per minute; when... When the speed approaches or exceeds 150 rpm, it indicates a sudden and significant drop in speed or overshoot, posing a risk of blockage or slippage in the conveying process. Take 0.

[0073] For the torque-speed coherence coefficient Since the larger the value, the more ordered it is, the forward normalization is directly applied without any additional processing, and the original value is retained. .

[0074] Determine the weight coefficients of the weighted fusion function. In this embodiment, the harvest orderliness index is denoted as... The calculation formula is: in, The weight of the normalized value of torque energy ratio. The weights for the normalized values ​​of the speed fluctuation amplitude. The weights of the normalized values ​​of the coherence coefficients satisfy the following conditions: And all weights are non-negative.

[0075] In this embodiment, the weight coefficients are determined using either the analytic hierarchy process (AHP) or a multi-objective optimization method based on field test data. First, several agricultural equipment experts are invited to conduct pairwise comparisons and scoring of the sensitivity of each characteristic parameter to the harvest blockage status, constructing a judgment matrix. The eigenvector corresponding to the largest eigenvalue of the judgment matrix is ​​calculated, and after passing a consistency check, the initial values ​​of each weight coefficient are obtained, rounded, and set to [value missing]. , , .

[0076] The rationale for this weighting is that the proportion of mid-to-high frequency energy in the torque signal is most sensitive to the instantaneous stall response, and it has the highest correlation with the final harvest effect in harvesting trials of multiple leafy vegetable varieties, so it is given the largest weight; the speed fluctuation amplitude reflects the cumulative effect of load changes, and its sensitivity is second; the coherence coefficient has a complementary role in identifying nonlinear anomalies in the system, and its weight is slightly lower but still retains a large influence.

[0077] Step 4: Using the harvest orderliness index as the optimization target and the allowable working pose space as the constraint, run a multivariable closed-loop optimization controller; the optimization controller adjusts the desired pitch angle and desired ground clearance of the header by continuously outputting perturbation signals, and automatically searches for the optimal header pose that keeps the harvest orderliness index at the target level based on the real-time response change trend of the harvest orderliness index.

[0078] In this embodiment, the multivariable closed-loop optimization controller adopts an online optimization architecture based on extreme value search control. The core idea of ​​this controller is to treat the harvesting process as a black box system where the mapping relationship between inputs (cropper pitch angle and ground clearance) and outputs (harvest orderliness index) is unknown and slowly time-varying. By injecting small trial signals and observing the output response, the gradient direction of the performance index is estimated online, thereby driving the header pose to iteratively converge towards the maximum value of the harvest orderliness index. In this way, without relying on a precise mathematical model of crop-machine interaction, it can adaptively cope with real-time changes in crop density, soil conditions, and terrain undulations.

[0079] In this embodiment, the workflow of the multivariable closed-loop optimization controller is as follows: Set a target threshold or target range for the harvest orderliness index, and superimpose a sine wave or pseudo-random wave of preset amplitude as a tentative pose command based on the expected pitch angle and expected ground clearance of the current header.

[0080] The method for setting the target threshold for the harvest orderliness index is as follows: based on the historical harvest data of this leafy vegetable variety under normal operating conditions, the harvest orderliness index is calculated. Based on the distribution characteristics, the 75th percentile value is taken as the target threshold. For example, after conducting a pre-harvest experiment on a 5-acre plot of a certain leafy vegetable, the data at each sampling time was statistically analyzed. The 75th percentile is 0.85. Set to 0.85. If using the target interval method, then set the target interval to... ,in, To allow for deviation, this embodiment uses 0.05, meaning the target interval is [0.80, 1]. By using the upper quantile of the statistical distribution as the target, we can ensure that the harvest quality is close to the historical best level, while also preserving a reasonable working margin for the optimization controller, avoiding frequent system adjustments or instability caused by pursuing overly demanding ideal values.

[0081] The specific method for superimposing tentative pose commands is as follows: At the current moment, let the current reference value of the desired pitch angle be... The current baseline value for the desired ground clearance is By superimposing sinusoidal perturbation signals, tentative pose commands are generated: in, The desired pitch angle command to be output at time t. The desired ground clearance command output at time t, in degrees and millimeters respectively; The amplitude of the pitch angle perturbation signal. The amplitude of the perturbation signal at altitude above ground; The angular frequency of the pitch angle perturbation signal. ω represents the angular frequency of the perturbation signal at altitude above ground; t represents the time variable.

[0082] The method for determining the perturbation amplitude is as follows: The adjustable range of the header pitch angle is set to 2% to 5%. In this embodiment, the adjustable range is -15° to +15°, with an amplitude of 0.5°. This value provides a sufficient signal-to-noise ratio response change without causing perceptible header vibration or reducing harvest quality due to excessive perturbations. Similarly, The adjustable range of ground clearance is 2% to 5%. In this embodiment, the adjustable range of ground clearance is -100 mm to +200 mm, and the amplitude is 8 mm.

[0083] The method for determining the perturbation angular frequency is as follows: and Different frequency values ​​are selected to achieve decoupling and identification of the two channels. In this embodiment... Set as Radius per second, or 0.2 Hertz; Set as Radius per second, or 0.3 Hz. The rationale for choosing different frequencies lies in: through... Spectral analysis of the response can separate the contributions of pitch angle changes and ground clearance changes to the harvest orderliness index in the frequency domain, avoiding crosstalk between the gradient estimates of the two channels. The selection of the frequency value balances system response speed and the discernibility of the complete signal cycle: if the frequency is too low, the gradient estimation will update slowly and converge slowly; if the frequency is too high, it may exceed the response bandwidth of the cutter head electro-hydraulic servo system, causing the perturbation signal to be filtered by the actuator and unable to be effectively transmitted to the harvesting process.

[0084] Execute the tentative pose command, and perform correlation analysis between the superimposed sine wave or pseudo-random wave in the tentative pose command and the real-time response of the monitored harvest orderliness index; determine the gradient direction of the harvest orderliness index with respect to the header pitch angle and ground clearance by comparing the phase relationship between the two, and determine the gradient magnitude by the ratio of the response amplitude to the amplitude of the superimposed sine wave or pseudo-random wave, thereby constructing a complete performance index gradient vector.

[0085] Gradient vector estimation employs a signal processing method based on multiplication demodulation, performed separately for the pitch angle channel and the ground clearance channel: For the pitch angle channel, the harvest orderliness index at time t will be collected in real time. Multiplied by frequency Same frequency sine wave The DC component is extracted using a low-pass filter to obtain... The gradient estimate of the pitch angle is denoted as... : in, This indicates a low-pass filter operation; the cutoff frequency of the low-pass filter is set to 0.05 Hz, far below the perturbation frequency of 0.2 Hz, to ensure effective filtering of high-frequency noise and intermodulation components, and extraction of the DC component representing gradient information. Factor This is the amplitude normalization coefficient, used to compensate for the influence of perturbation signal amplitude on gradient amplitude estimation.

[0086] Similarly, for passageways at ground level, Multiplied by frequency Same frequency sine wave The DC component is extracted using a low-pass filter with the same cutoff frequency to obtain... The gradient estimate of the height above the ground is denoted as . : Will and Combining these elements to form a two-dimensional performance index gradient vector. The direction of the gradient vector G indicates the direction of adjustment in which the exponential increase in orderliness is fastest at the current pose; The sign of the value determines whether the pitch angle should be adjusted to increase (tilt down) or decrease (tilt up). The sign of the value determines whether the height above the ground should be adjusted to increase or decrease.

[0087] For example, in a certain optimization cycle, if the demodulation calculation yields... , This indicates that, based on the current position, appropriately increasing the pitch angle (making the cutter closer to horizontal or slightly lowered) and appropriately reducing the height above the ground is expected to improve the harvest orderliness index.

[0088] Along the gradient vector direction, with an adaptive step size positively correlated with the gradient magnitude, the desired pitch angle and desired ground clearance are updated to generate a new cutter pose command; this new cutter pose command is taken as the optimal cutter pose obtained in this optimization. The pose update is performed according to the following adaptive step size formula: in, This represents the new desired pitch angle generated after this round of optimization and update, in degrees. It is the pitch angle command value to be output to the decoupled controller in the next step. This represents the new expected ground clearance, in millimeters, generated after this round of optimization and update. It is the ground clearance command value to be output to the decoupled controller in the next step. For symbolic functions, This is a preset reference value for the normalized gradient magnitude.

[0089] parameter The process for determining the values ​​is as follows: Before the harvesting operation officially begins, an offline calibration program is executed. Specifically, the offline calibration program involves: running the harvester at a constant speed without load on the field to be harvested (the conveying components do not perform actual conveying operations); within the adjustable range of the header pitch angle and ground clearance, traversing all pose combinations with a preset step size; and at each pose combination, applying the aforementioned amplitude to the header pitch angle and ground clearance. and The sinusoidal perturbation signal was used to simultaneously record the harvest order index. The response is used to calculate the gradient estimate at each sampling pose using the multiplication demodulation method described in step 4. and This generates a two-dimensional gradient vector, and its Euclidean norm is calculated. After traversal, the average of the gradient norms of all sampled pose points is taken as the mean. The value of .

[0090] This embodiment further clarifies the step size update strategy as follows: Let the pitch angle adjustment step size be denoted as... The step size for adjusting the height off the ground is recorded as follows: The calculation formula is: in, As a reference step size, the pitch angle reference step size in this embodiment is 0.3°, and the ground clearance reference step size is 5 mm; , where is the Euclidean norm of the gradient vector, representing the overall steepness of the performance index change at the current pose point; Let be a very small positive number, take This is used to avoid the denominator being zero.

[0091] Updated expected pitch angle and desired ground clearance for: The rationale for using this adaptive step size formula is as follows: when the gradient magnitude is large, it indicates that the current pose deviates far from the optimal pose, and a larger step size can accelerate convergence by quickly approximating the target pose; when the gradient magnitude is small, it indicates that the target point is approaching, and automatically reducing the step size helps to reduce steady-state oscillations and improve optimization accuracy. Reference step size. The value is determined by taking into account the adjustment resolution of the cutting table's mechanical structure (0.1° pitch angle and 2 mm ground clearance) and the step response time of the electro-hydraulic servo system (approximately 0.5 seconds to reach 90% of steady state). The maximum allowable adjustment within one optimization cycle is set to 3 to 5 times the adjustment resolution, which ensures that the pose can be effectively updated while avoiding overshoot or oscillation caused by excessive adjustment in a single operation.

[0092] Before outputting the new cutting table pose command, verify whether it falls within the allowed working pose space. If it exceeds the safety boundary, project it to the nearest point on the boundary of the allowed working pose space.

[0093] The safety boundary projection operation is described as follows: In the allowed job pose space constructed in step 2, retrieve the point that matches the pose to be output. The feasible pose point with the minimum Euclidean distance. The retrieval method can employ a nearest neighbor search algorithm: store the allowed job pose space in the form of a lookup table, traverse each feasible pose point in the table, and calculate its distance from the nearest neighbor. The distances are calculated, and the point with the smallest distance is chosen as the projection point. ,by As the output of the corrected cutter pose command. If If the cutting table is already within the permissible working pose space, no correction is needed, and the output can be performed directly. This constraint projection step ensures that the cutting table will not exceed the safety boundary defined by the minimum cutting depth value or the damage limit cutting depth value under any circumstances, providing a hard safety guarantee from the control level.

[0094] The above process is continuously repeated to form a closed-loop optimization control with the harvest orderliness index as the optimization objective. The execution cycle of the optimization controller is set to 5 seconds. This execution cycle is determined based on the fact that each cycle contains at least 2 to 3 complete perturbation signal cycles to ensure the accuracy of gradient demodulation. Calculated at a low perturbation frequency of 0.2 Hz, a single signal cycle is 5 seconds, and collecting 10 seconds of data in each optimization cycle can cover 2 complete cycles, meeting the demodulation accuracy requirements. At the same time, the 5-second cycle matches the update frequency of the predicted surface elevation curve in step 1, ensuring the synergy between pose optimization and terrain prediction.

[0095] In this embodiment, when the multivariate closed-loop optimization controller fails to make the harvest orderliness index reach the target threshold or target range within a preset time period, the system triggers a graded response mechanism; the mechanism includes: issuing a warning to the operator, reducing the harvester's travel speed, or performing a safe shutdown.

[0096] The triggering conditions and execution logic of the tiered response mechanism are defined as follows: Define the timeout threshold. It takes 60 seconds, which is 12 consecutive optimization cycles.

[0097] Level 1 response: When Within 6 consecutive optimization cycles (i.e., 30 seconds), it was lower than If the temperature does not rise, it is considered a minor anomaly, triggering a warning to the operator. The warning is triggered by an intermittent beeping of the buzzer in the cab, while the human-machine interface displays the text message "Harvesting orderliness is low; it is recommended to check the feeding status."

[0098] Level 2 response: When It was lower than the target value for nine consecutive optimization cycles (i.e., 45 seconds). If the speed continues to rise without any upward trend, it is considered a moderate anomaly, triggering a reduction in the harvester's travel speed. The speed reduction method involves sending a travel speed command to the travel drive controller at 70% of the current value. The rationale for the speed reduction is that it reduces the amount of crop fed per unit time, lightens the load on the conveying components, creates conditions for restoring orderly conveying, and simultaneously maintains uninterrupted harvesting operations.

[0099] Level 3 response: When It was lower than the target value for 12 consecutive optimization cycles (i.e., 60 seconds). or at any time When the index drops sharply below 0.1, it is considered a serious anomaly, triggering a safety shutdown. The shutdown procedure involves stopping the walking drive and disconnecting the power output to the cutter and conveyor components. The rationale for this response is that prolonged low harvest order or a sudden drop in the index indicates that the conveyor channel may have experienced severe blockage or mechanical failure. Continuing to operate would not only compromise harvest quality but could also cause mechanical damage to the conveyor components or overload and burn out the drive motor.

[0100] Step 5: Input the optimal cutter table pose into the preset decoupled controller based on the inverse dynamics model to generate mutually decoupled control commands, which independently drive the cutter table pitch adjustment mechanism and the lifting adjustment mechanism to achieve real-time control of the cutter table pose.

[0101] In this embodiment, the decoupling controller embeds an inverse dynamics model that characterizes the dynamic coupling relationship between the pitch and rise motion of the cutter table. The inverse dynamics model takes the optimal cutter table pose as input and calculates the feedforward control command to compensate for the cross-coupling effect of the two sets of hydraulic actuators, so as to drive the actual pose of the cutter table to track the optimal cutter table pose within the preset overshoot allowable range.

[0102] The header pitch adjustment mechanism includes at least one pitch hydraulic cylinder, and the lifting adjustment mechanism includes at least one lifting hydraulic cylinder. The pitch and lifting hydraulic cylinders are each driven by an independent high-speed response electro-hydraulic servo valve. The header is connected to the harvester chassis via a spatial linkage mechanism, which creates an inherent kinematic and dynamic coupling effect between the pitch and lifting movements: driving the pitch hydraulic cylinder alone not only changes the header pitch angle but also causes a linked change in the header's ground clearance; conversely, driving the lifting hydraulic cylinder alone also interferes with the header pitch angle. Without compensation for this coupling effect, the tracking accuracy of the header's actual posture will decrease, the adjustment time will increase, and it may even induce oscillations between the pitch and lifting channels. This embodiment introduces a decoupling controller based on an inverse dynamics model. Its purpose is to actively compensate for the aforementioned cross-coupling effect through a feedforward approach, enabling the pitch and lifting channels to be controlled according to their independent dynamic characteristics, achieving a decoupling control effect without interference.

[0103] In this embodiment, the specific construction and solution method of the inverse dynamics model is as follows: A positive dynamic model of the electro-hydraulic servo system of the cutting platform is established. This positive dynamic model uses the pitch angle and ground clearance of the cutting platform as generalized coordinates. It describes the dynamic behavior of the cutting platform under the simultaneous drive of the pitch adjustment mechanism and the height adjustment mechanism through the inertia matrix, the Coriolis force and centrifugal force matrix, and the gravity vector. The off-diagonal elements of the inertia matrix and the Coriolis force and centrifugal force matrix represent the cross-coupling effect of the motion of the pitch adjustment mechanism and the height adjustment mechanism. The input of the positive dynamic model is the driving force and the driving torque, and the output is the acceleration of the ground clearance and pitch angle of the cutting platform.

[0104] Specifically, a positive dynamic model is established using the Lagrange equations: a generalized coordinate vector is defined. ,in, The height of the cutting platform above the ground, in meters; Let be the pitch angle of the cutting table, in radians. The Lagrangian function L of the cutting table system is the difference between the system's kinetic energy T and potential energy V, i.e., L = TV. The kinetic energy T consists of the translational kinetic energy and rotational kinetic energy about the center of mass of the cutting table and its components (including the conveyor components, cutter assembly, etc.); the potential energy V consists of gravitational potential energy and the compressive potential energy of the hydraulic oil in the hydraulic cylinders. The standard form of the positive dynamic model is derived as follows: in, For generalized velocity vectors, It is a generalized acceleration vector; The inertia matrix is ​​a 2×2 symmetric positive definite matrix. The matrix represents the Coriolis force and centrifugal force in a 2×2 form. It is a 2×1 gravity vector; It is a 2×1 generalized force vector. , For equivalent force acting on the generalized coordinate system of altitude above the ground, the unit is Newton. The equivalent torque acting on the generalized coordinates of the pitch angle is expressed in Newton-meters.

[0105] Inertia matrix The specific expression is: in, The equivalent inertia related to translation in the direction of height above the ground, The equivalent moment of inertia of rotation about the pitch axis. This is the coupled inertia between translation and rotation. and Because the center of mass of the cutting platform is not on the pitch axis, the pitch motion causes a change in the vertical displacement of the center of mass, which in turn creates a dynamic coupling with the lifting motion.

[0106] Coriolis force and centrifugal force matrix The specific expression is: Each element is a function of the generalized coordinates and their derivatives; off-diagonal elements and The equivalent Coriolis force in the height-to-ground direction caused by the pitch velocity of the cutter, and the equivalent torque in the pitch direction caused by the change in height-to-ground direction, represent another level of coupling source.

[0107] Gravity vector ,in, The contribution of gravity to the generalized coordinates of height above the ground. This represents the contribution of gravity to the generalized coordinates of the pitch angle. When the cutter is at different pitch angles, the distribution of the gravity components on the two generalized coordinates changes, further demonstrating the coupling effect.

[0108] The model parameters are obtained as follows: the three-dimensional model of the cutting table and its auxiliary components is read by computer-aided design software, and inertial parameters such as mass, center of mass position and moment of inertia about each axis are extracted; for parameters such as hydraulic cylinder viscous damping that are difficult to obtain accurately, a system identification experiment is used for calibration, that is, a known sinusoidal sweep force or step force is applied to a single hydraulic cylinder in the bench test, the corresponding generalized coordinate response is measured, and the undetermined coefficients in the model are obtained by fitting with the least squares method.

[0109] The inverse dynamic model is obtained by mathematically inverting the forward dynamic model, as follows: Based on the positive dynamics model, the generalized acceleration vector is... Treating it as known input, the generalized force vector can be directly solved. ,Right now: in, Let the desired generalized acceleration vector be... ; Let the desired generalized velocity vector be... ;here This represents the current actual generalized coordinate vector. The inverse dynamics model uses the desired pose motion parameters. and The input is the generalized force vector required to achieve the desired motion, and the output is the generalized force vector. Since matrices M, C, and vector G all contain coupling terms between the pitch and climb channels, the inverse dynamics model automatically compensates for these coupling effects when calculating the driving force and driving torque.

[0110] The optimal cutter pose, along with the velocity and acceleration calculated from this optimal cutter pose through differential or filtering, are substituted into the inverse dynamics model to calculate and output the corresponding desired driving force and desired driving torque. The optimal cutter pose... The output of the optimization controller from step 4. In practical discrete control systems, the differential processing of the optimal cutter pose signal is performed using the following method: Desired altitude and speed and desired pitch angular velocity Obtained through first-order backward difference combined with low-pass filtering: in, This represents the sequence number of the discrete control cycle. To control the period, this embodiment uses 10 milliseconds; the LPF is a first-order low-pass filter with a cutoff frequency set to 30 Hz, used to suppress high-frequency noise amplification introduced by differential operations.

[0111] Desired acceleration above ground and desired pitch angle acceleration The speed signal was obtained by performing a first-order backward differential and low-pass filtering on the speed signal again, with the same filter cutoff frequency of 30 Hz.

[0112] Substituting the aforementioned pose, velocity, and acceleration signals into the inverse dynamics model, the desired equivalent driving force within this control cycle can be calculated. and expected equivalent moment .

[0113] Based on the pre-calibrated hydraulic cylinder force-pressure mapping relationship and the flow-pressure-current characteristics of the electro-hydraulic servo valve, the desired driving force and desired driving torque are sequentially converted into the target driving pressure of each hydraulic cylinder, and then converted into the feedforward control current signal to drive the servo valve. This feedforward control current signal is the feedforward control command generated by the decoupling controller.

[0114] The conversion of generalized force into the driving force of each hydraulic cylinder is achieved through the force Jacobian matrix: Let the driving force of the pitch hydraulic cylinder be... The driving force of the lifting hydraulic cylinder is The vector formed by the two is Generalized force vector With hydraulic cylinder driving force The relationship between them is: in, It is a 2×2 force-jacobi matrix, whose elements are the transmission ratios between the extension and retraction velocities of each hydraulic cylinder and the corresponding generalized coordinate velocities. This force-jacobi matrix changes with the current pose of the cutting table. It changes with the changes. The specific expression is obtained by kinematic analysis of the geometric dimensions of the cutting table linkage mechanism and the coordinates of the hinge points of each hydraulic cylinder.

[0115] Inverting this relation yields the conversion formula from generalized force to hydraulic cylinder driving force: in, This is the expected generalized force vector output by the inverse dynamics model.

[0116] The target driving pressure of the hydraulic cylinder is calculated by the following formula: in, This indicates the target driving pressure of the lifting hydraulic cylinder; This indicates the target driving pressure of the pitch hydraulic cylinder; This indicates the driving force output by the lifting hydraulic cylinder; This indicates the driving force output by the pitch hydraulic cylinder; The effective working area of ​​the lifting hydraulic cylinder. The effective working area of ​​the pitching hydraulic cylinder, in square meters, is calculated from the cylinder diameter and piston rod diameter parameters.

[0117] The feedforward control current signal of the electro-hydraulic servo valve is determined based on the servo valve's pressure-current characteristic curve. This characteristic curve is provided by the servo valve manufacturer or obtained through calibration on a hydraulic test bench, describing the steady-state mapping relationship between the valve output pressure and the control current under a given supply pressure. In this embodiment, the -10V to +10V control signal of the servo valve corresponds to -100% to +100% of the valve spool opening, and the feedforward control current signal... and The target driving pressure is obtained by either table lookup or linear interpolation. and The mapping is obtained.

[0118] This embodiment conducted a complete harvesting operation test on a standard experimental field. The experimental conditions were as follows: the leafy vegetable variety was Shanghai dwarf bok choy, with a plant height of approximately 22 cm. The planting ridges had artificially created continuous undulating sections (amplitude ±6 cm, wavelength approximately 1.5 m). The harvester operated at a constant travel speed of 0.8 m / s. During the experiment, the multivariate closed-loop optimization controller was activated throughout, and the target threshold for the harvest orderliness index was set. The value was set to 0.85, and the optimization period was set to 5 seconds. The experiment lasted for 120 seconds, with one set of data recorded every 3 seconds, for a total of 40 sets. The recorded parameters included: harvest orderliness index. The height of the cutting table from the ground and the torque of the drive motor are shown in the table below: Table 1. Example of data from the harvesting operation trial: Based on Table 1 above, Figure 2 and Figure 3 As shown, the entire operation process can be divided into five typical stages, each with different data characteristics and the technical effects it represents:

[0119] Phase 1 (0 (24s, normal operation phase): The ground surface is flat, and the height of the cutting platform above the ground is stable at 73. Within a 76mm range, the drive motor torque remains at 11.5. Within a low fluctuation range of 14.8 Nm. Harvest order index. At 0.88 The index fluctuated slightly between 0.94 and 0.91, with a mean of 0.91, consistently above the target threshold of 0.85. This stage indicates that, under suitable header orientation, the load on the conveying components is stable, the proportion of mid-to-high frequency energy in the torque signal is low, the speed fluctuation is small, and the harvest orderliness assessment model can accurately reflect the orderly harvesting state, with the output index remaining stable in the high range.

[0120] Phase Two (24) (45s, the stage of index decline due to terrain protrusion): As the harvester enters a section of continuous terrain protrusion, the height of the header passively decreases from 74mm to 60mm due to the rising terrain, corresponding to a gradual increase in the cutting depth of the cutter. The drive motor torque continuously climbs from 14.8Nm to 32.4Nm, indicating that the increased soil content at the root and the rise in conveying resistance are caused by the increased cutting depth. The harvesting orderliness index drops from 0.84 to 0.58, falling below 0.58 for the first time at 45s. The secondary early warning threshold (0.7) was verified in this stage. This stage verifies that the predicted surface elevation curve constructed in this embodiment can detect terrain changes in advance, but it also shows that simply relying on terrain feedforward information is not enough to cope with the dynamic changes of complex terrain. It is necessary to introduce a closed-loop control mechanism with harvest status as feedback.

[0121] Phase 3 (48) (60s, Blockage Worsening Stage): The height of the cutting platform above the ground continues to decrease to 55mm at 49s and 49mm at 60s, reaching the critical range of the damage limit cutting depth set in step 2. The torque of the drive motor rises sharply to 68.5Nm at 60s, approximately five times the normal value. The harvest orderliness index declines rapidly, reaching a low of 0.23 at 60s. This stage verifies the rationality of using the time-frequency characteristics of the torque signal as a parameter representing the harvest state in step 3 of this embodiment. That is, when the amount of soil carried in the conveying channel increases dramatically due to excessive cutting depth, the sudden change in the load of the drive motor will be significantly reflected in both the time domain (torque amplitude) and the frequency domain (proportion of mid-to-high frequency energy). The weighted and fused harvest orderliness index further condenses the multidimensional physical signal changes to 0. 1. Scalar metrics provide the controller with clear and sensitive optimization targets.

[0122] Phase Four (63) (84s, optimization controller adjustment phase): After the harvest order index reaches a trough of 0.23, the multivariate closed-loop optimization controller starts active adjustment. Figure 2 The height of the cutting platform from the ground gradually increased from 49mm to 76mm, with the recovery process showing a characteristic of being slow at first and then fast: 63 72s corresponds to the controller's trial-and-error process of continuously outputting sinusoidal perturbation signals and estimating the gradient of performance indicators online. Due to the weak initial gradient information, the pose adjustment range is relatively conservative; 72 At 84s, as the gradient estimation direction gradually becomes clear, the controller rapidly advances along the gradient direction with an adaptive step size. Correspondingly, Figure 3 The torque of the drive motor continuously decreased from 68.5 Nm to 17.8 Nm; the harvest orderliness index steadily increased from 0.26 to 0.8. This stage also verified the two core functions of the multivariable closed-loop optimization controller: by continuously outputting perturbation signals and demodulating them in the frequency domain, it can effectively identify the gradient direction of the influence of the header pitch angle and ground clearance on the harvest orderliness without relying on a precise mathematical model; the adaptive step size update strategy along the gradient direction ensures that the pose adjustment process avoids oscillations caused by blind large-scale adjustments and guarantees sufficient convergence speed when gradient information is sufficient.

[0123] Phase 5 (87) 120s, recovery and stabilization phase): From 87s onwards, It regained the target threshold of 0.85, and remained at 0.85 thereafter. The torque fluctuated stably within the 0.93 range, and the drive motor torque recovered and remained at 11.3. The normal level is 14.0 Nm, and the height of the cutting platform above the ground is stable at 74. 76mm. This stage indicates that the optimization controller successfully guided the cutting table to the optimal working point within the allowed working position space, and the conveying process returned to an orderly state.

[0124] Please see Figure 4 The present invention also provides an electro-hydraulic control device for the height of the header of a leafy vegetable harvester. This device is used to implement the aforementioned electro-hydraulic control method for the height of the header of a leafy vegetable harvester, comprising: The surface topography prediction module is used to collect real-time data on the aircraft's pitch angle, side tilt angle, and the three-dimensional point cloud of the surface of the area to be harvested. After spatiotemporal alignment and fusion, it constructs a predicted surface elevation profile based on the current position of the cutter and including future time series. The safety boundary construction module is used to set layered control thresholds, including the minimum cutting depth to ensure the cutting of vegetable roots, the optimal cutting depth range to achieve orderly transportation and minimal soil carrying, and the damage limit cutting depth to prevent damage to the vegetable heart; the layered control thresholds are combined with the predicted ground elevation profile to determine the allowable operating posture space of the cutting platform within the operating width; the allowable operating posture space defines the safety boundaries of the cutting platform's pitch angle and height above the ground; The harvest orderliness assessment module is used to collect torque and speed signals of the drive motor of the conveying component in real time during the harvesting operation, perform time-frequency analysis, extract characteristic parameters that characterize the harvesting state, and calculate the harvest orderliness index at the current moment through a preset harvest orderliness assessment model. The pose closed-loop optimization module is used to run a multivariable closed-loop optimization controller with the harvest orderliness index as the optimization target and the allowable working pose space as the constraint. The optimization controller adjusts the desired pitch angle and desired ground clearance of the header by continuously outputting perturbation signals, and automatically searches for the optimal header pose that keeps the harvest orderliness index at the target level based on the real-time response change trend of the harvest orderliness index. The decoupled execution control module is used to input the optimal cutter table posture into a preset decoupled controller based on an inverse dynamics model, generate mutually decoupled control commands, and independently drive the cutter table pitch adjustment mechanism and the lifting adjustment mechanism to achieve real-time control of the cutter table posture.

[0125] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0126] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0127] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0128] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for electro-hydraulic control of the header height of a leafy vegetable harvester, characterized in that, The specific steps include: Step 1: Real-time acquisition of aircraft pitch angle, roll angle data and 3D point cloud of the ground surface of the area to be harvested; after spatiotemporal alignment and fusion, construct a predicted ground elevation profile with the current position of the cutter as the reference and including future time series. Step 2: Set layered control thresholds, including the minimum cutting depth to ensure the cutting of vegetable roots, the optimal cutting depth range to achieve orderly transport and minimal soil load, and the damage limit cutting depth to prevent damage to the vegetable heart; combine the layered control thresholds with the predicted ground elevation profile to determine the allowable working posture space of the cutting platform within the working width; the allowable working posture space defines the safety boundaries of the cutting platform's pitch angle and height above the ground. Step 3: During the harvesting operation, the torque and speed signals of the drive motor of the conveying component are collected in real time and analyzed in time and frequency. Characteristic parameters representing the harvesting state are extracted, and the harvesting orderliness index at the current moment is calculated through the preset harvesting orderliness evaluation model. Step 4: Using the harvest orderliness index as the optimization target and the allowable working pose space as the constraint, run a multivariable closed-loop optimization controller; the optimization controller adjusts the desired pitch angle and desired ground clearance of the header by continuously outputting perturbation signals, and automatically searches for the optimal header pose that keeps the harvest orderliness index at the target level based on the real-time response change trend of the harvest orderliness index. Step 5: Input the optimal cutter table pose into the preset decoupled controller based on the inverse dynamics model to generate mutually decoupled control commands, which independently drive the cutter table pitch adjustment mechanism and the lifting adjustment mechanism to achieve real-time control of the cutter table pose.

2. The electro-hydraulic control method for the header height of a leafy vegetable harvester according to claim 1, characterized in that, During the spatiotemporal alignment stage, the timestamp of each frame of the 3D point cloud on the ground is paired with the pitch and roll angles of the machine in the same frame. Based on the harvester's travel speed, the point clouds at different times are transformed into the vehicle coordinate system with the current position of the cutter as the origin to obtain the pose-corrected point cloud data. The pose is defined by the pitch angle of the cutter and the height above the ground. The corrected point cloud data within the preset time window are stitched together and interpolated to form a predicted surface elevation curve covering the entire cutting platform operation width; this predicted surface elevation curve reflects the surface undulation and reflects the elevation difference between the rows to be harvested on the transverse profile.

3. The electro-hydraulic control method for the header height of a leafy vegetable harvester according to claim 2, characterized in that, The specific settings for the hierarchical control thresholds are as follows: Based on the transverse profile of the predicted ground elevation surface at the current position of the cutter, and combined with the leafy vegetable variety, growth height and root depth, the minimum cutting depth value is determined; this minimum cutting depth value ensures that the actual cutting depth at any transverse position is greater than the preset allowable root residual length. Above the minimum cutting depth, based on the upper limit of the amount of soil attached to the roots and the requirement for stem integrity required for orderly conveying, the lower and upper limits of the optimal cutting depth range are determined so that when the cutter operates within this range, it not only cuts the vegetable roots but also ensures that the amount of soil attached does not exceed the preset threshold, while also ensuring that the clamping position of the conveying component on the leafy vegetables is in the middle of the stem. The damage limit cutting depth value is determined based on the height of the cabbage from the ground and the anti-blocking safety gap of the conveying channel; when the actual cutting depth reaches or exceeds the damage limit cutting depth value, it is determined that there is a risk of cutting the cabbage or blocking it.

4. The electro-hydraulic control method for the header height of a leafy vegetable harvester according to claim 3, characterized in that, The construction of the permitted job pose space is specifically as follows: On the predicted ground elevation surface, all combinations of cutter pitch angles and ground clearance are traversed within the working width; combinations that would cause the predicted cut depth at any lateral position to exceed the damage limit cut depth value or fall below the minimum cut depth value are eliminated; the remaining combinations are defined as the allowed working pose space.

5. The electro-hydraulic control method for the header height of a leafy vegetable harvester according to claim 1, characterized in that, The construction and calculation of the harvest orderliness assessment model specifically includes: The characteristic parameters include: the energy ratio of the drive motor torque signal within a preset frequency band, the low-frequency fluctuation amplitude of the speed signal, and the coherence coefficient between the torque and speed signals; The harvest orderliness evaluation model is a weighted fusion function or a pre-trained machine learning model. Its input is the feature parameters calculated in real time, and its output is a harvest orderliness index between 0 and 1. Here, 1 represents an ideal ordered state, and 0 represents a completely blocked or severely abnormal state.

6. The electro-hydraulic control method for the header height of a leafy vegetable harvester according to claim 1, characterized in that, The workflow of the multivariable closed-loop optimization controller is as follows: Set a target threshold or target range for the harvest orderliness index, and superimpose a sine wave or pseudo-random wave of preset amplitude as a tentative pose command based on the expected pitch angle and expected ground height of the current header. Execute the tentative pose command, and perform correlation analysis between the superimposed sine wave or pseudo-random wave in the tentative pose command and the real-time response of the monitored harvest orderliness index; determine the gradient direction of the harvest orderliness index with respect to the header pitch angle and ground clearance by comparing the phase relationship between the two, and determine the gradient magnitude by the ratio of the response amplitude to the amplitude of the superimposed sine wave or pseudo-random wave, thereby forming a complete performance index gradient vector; Along the gradient vector direction, with an adaptive step size positively correlated with the gradient magnitude, update the desired pitch angle and desired ground clearance to generate a new cutter pose command; the new cutter pose command is the optimal cutter pose obtained in this optimization. Before outputting the new cutting table pose command, verify whether it falls within the allowed working pose space. If it exceeds the safety boundary, project it to the nearest point on the boundary of the allowed working pose space. The above process is continuously executed in a loop to form a closed-loop optimization control with the harvest orderliness index as the optimization objective.

7. The electro-hydraulic control method for the height of the header of a leafy vegetable harvester according to claim 6, characterized in that, When the multivariate closed-loop optimization controller fails to make the harvest orderliness index reach the target threshold or target range within a preset time period, the system triggers a graded response mechanism; the mechanism includes: issuing a warning to the operator, reducing the harvester's travel speed, or performing a safe shutdown.

8. The electro-hydraulic control method for the header height of a leafy vegetable harvester according to claim 1, characterized in that, The decoupling controller is embedded with an inverse dynamics model that characterizes the dynamic coupling relationship between the pitch and rise of the cutter head. The inverse dynamics model takes the optimal cutter head pose as input and calculates the feedforward control command to compensate for the cross-coupling effect of the two sets of hydraulic actuators, so as to drive the actual pose of the cutter head to track the optimal cutter head pose within the preset overshoot allowable range.

9. The electro-hydraulic control method for the height of the header of a leafy vegetable harvester according to claim 8, characterized in that, The specific construction and solution method of the inverse dynamics model is as follows: A positive dynamic model of the electro-hydraulic servo system of the cutting platform is established. The positive dynamic model uses the pitch angle and ground clearance of the cutting platform as generalized coordinates. It describes the dynamic behavior of the cutting platform under the simultaneous drive of the pitch adjustment mechanism and the lifting adjustment mechanism through the inertial matrix, the Coriolis force and centrifugal force matrix and the gravity vector. The off-diagonal elements of the inertial matrix and the Coriolis force and centrifugal force matrix represent the cross-coupling effect of the motion of the pitch adjustment mechanism and the lifting adjustment mechanism. The positive dynamics model takes the driving force and driving torque as inputs and outputs the acceleration of the cutter's height above the ground and pitch angle. The positive dynamics model is mathematically inverted to obtain the inverse dynamics model; the input of the inverse dynamics model is the desired cutter pose and the velocity and acceleration of the cutter pose, and the output is the desired driving force and desired driving torque required to achieve the desired cutter pose. The optimal cutter pose, along with the velocity and acceleration calculated from the optimal cutter pose through differential or filtering, are substituted into the inverse dynamics model to calculate and output the corresponding desired driving force and desired driving torque. Based on the pre-calibrated hydraulic cylinder force-pressure mapping relationship and the flow-pressure-current characteristics of the electro-hydraulic servo valve, the desired driving force and desired driving torque are sequentially converted into the target driving pressure of each hydraulic cylinder, and then converted into the feedforward control current signal to drive the servo valve. This feedforward control current signal is the feedforward control command generated by the decoupling controller.

10. A height control device for the header of a leafy vegetable harvester, characterized in that, The aforementioned electro-hydraulic control device for the height of the header of a leafy vegetable harvester is used to implement the electro-hydraulic control method for the height of the header of a leafy vegetable harvester as described in any one of claims 1-9, comprising: The surface topography prediction module is used to collect real-time data on the aircraft's pitch angle, side tilt angle, and the three-dimensional point cloud of the surface of the area to be harvested. After spatiotemporal alignment and fusion, it constructs a predicted surface elevation profile based on the current position of the cutter and including future time series. The safety boundary construction module is used to set layered control thresholds, including the minimum cutting depth to ensure the cutting of vegetable roots, the optimal cutting depth range to achieve orderly transportation and minimal soil carrying, and the damage limit cutting depth to prevent damage to the vegetable heart; the layered control thresholds are combined with the predicted ground elevation profile to determine the allowable operating posture space of the cutting platform within the operating width; the allowable operating posture space defines the safety boundaries of the cutting platform's pitch angle and height above the ground; The harvest orderliness assessment module is used to collect torque and speed signals of the drive motor of the conveying component in real time during the harvesting operation, perform time-frequency analysis, extract characteristic parameters that characterize the harvesting state, and calculate the harvest orderliness index at the current moment through a preset harvest orderliness assessment model. The pose closed-loop optimization module is used to run a multivariable closed-loop optimization controller with the harvest orderliness index as the optimization target and the allowable working pose space as the constraint. The optimization controller adjusts the desired pitch angle and desired ground clearance of the header by continuously outputting perturbation signals, and automatically searches for the optimal header pose that keeps the harvest orderliness index at the target level based on the real-time response change trend of the harvest orderliness index. The decoupled execution control module is used to input the optimal cutter table posture into a preset decoupled controller based on an inverse dynamics model, generate mutually decoupled control commands, and independently drive the cutter table pitch adjustment mechanism and the lifting adjustment mechanism to achieve real-time control of the cutter table posture.

Images