Intelligent planter for tobacco seedlings

Abstract

The application relates to the technical field of planting devices and discloses an intelligent planter for tobacco seedlings, which comprises a motor set cooperative control unit, a transplanting execution unit and a ridge surface planting posture optimization unit. The motor set cooperative control unit is used for taking the global time reference criterion established by a pulse coupling synchronization algorithm as a constraint condition, generating a driving motor set synchronization control instruction according to a whole machine control instruction and a communication addressing reference, and driving the motor set to run in a preset driving state. The transplanting execution unit is used for generating a cooperative action time sequence instruction matched with a seedling throwing mechanism, a duckbill mechanism and a driving state reference signal, driving the seedling throwing mechanism to execute a seedling throwing action and driving the duckbill mechanism to complete an adaptive profiling planting action in sequence according to the cooperative action time sequence instruction, and transmitting action execution and feedback data to the ridge surface planting posture optimization unit in real time during the execution of the seedling throwing and planting actions. The application effectively realizes accurate linkage matching of the driving state and the transplanting action, and has the abilities of advanced perception and adaptive compensation for the change of the ridge surface terrain and the soil resistance.

Description

Technical Field

[0001] This invention relates to the field of planting equipment technology, and more specifically, to an intelligent planter for tobacco seedlings. Background Technology

[0002] Tobacco is one of my country's major economic crops. From sowing and seedling cultivation to harvesting, it involves several key operational stages, among which the transplanting stage has long relied heavily on manual labor. With the advancement of domestic technology and the continuous upgrading of agricultural mechanization, accelerating the mechanization of tobacco transplanting has become an important way to improve operational efficiency, increase farmers' income, and ensure the sustainable development of the tobacco industry.

[0003] However, existing tobacco seedling planters on the market still have significant technical shortcomings, making it difficult to meet the needs of precise and stable transplanting operations. They cannot achieve precise phase alignment and synchronous collaborative control of multiple drive motors, and are prone to motor synchronization deviations under conditions of fluctuating field speed, load changes, and turning operations, directly causing disorder in the overall transplanting rhythm. At the same time, the seedling feeding mechanism and the duckbill mechanism are difficult to combine with the reference signal of the overall machine's driving status to generate dynamic collaborative timing instructions, and cannot adaptively match the seedling feeding frequency and planting action rhythm with the real-time driving speed. In actual operation, quality problems such as uneven plant spacing, missing seedlings, and overlapping seedlings are very likely to occur.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] In view of the problems in related technologies, this invention proposes an intelligent planter for tobacco seedlings to overcome the aforementioned technical problems existing in the existing related technologies.

[0006] Therefore, the specific technical solution adopted by the present invention is as follows: An intelligent planter for tobacco seedlings, the planter comprising: The whole machine control unit is used to integrate the pre-acquired whole machine operating condition control signals and safety interlock control signals into whole machine control commands, and send the whole machine control commands to the motor set cooperative control unit; The motor set coordination control unit is used to allocate communication station numbers to the drive motor set based on the extended communication protocol to establish a motor communication addressing reference. The global time reference criterion established by the pulse coupling synchronization algorithm is used as a constraint condition. The drive motor set synchronization control command is generated according to the whole machine control command and the communication addressing reference. The walking drive and steering unit is used as the driving power source of the drive motor group. Based on the synchronous control command of the drive motor group, it coordinates the operation status between each drive motor to perform driving and steering actions, and synchronously outputs the driving status reference signal to the transplanting execution unit. The transplanting execution unit is used to generate coordinated action timing instructions that match the seedling feeding mechanism, the duckbill mechanism and the driving status reference signal according to the preset driving speed-planting frequency matching rules. The unit drives the seedling feeding mechanism to perform seedling feeding action and drives the duckbill mechanism to cooperate in completing adaptive contour planting action according to the coordinated action timing instructions. The unit acquires action execution and feedback data in real time during the seedling feeding and planting actions and transmits it to the ridge planting posture optimization unit. The ridge planting posture optimization unit is used to analyze action execution and feedback data, generate ridge planting posture global adjustment commands, update the machine's walking, steering and planting matching operating parameters according to the ridge planting posture global adjustment commands, and send the updated operating parameters back to the machine control unit.

[0007] Preferably, the ridge planting posture adjustment command includes: a guiding driving action command along both sides of the ridge, an adjustment action command to adapt to the height and parallelism of the ridge, an adjustment action command to match the width of the ridge, and an adjustment action command to adjust the spacing between seedlings.

[0008] Preferably, the motor set collaborative control unit, when assigning communication station numbers to the drive motor set based on the extended communication protocol to establish a motor communication addressing reference, and using the global time reference criterion established by the pulse coupling synchronization algorithm as a constraint condition, generates synchronous control commands for the drive motor set according to the overall machine control commands and the communication addressing reference, includes: The extended communication protocol is configured using a dual verification mechanism that combines parity check and cyclic redundancy check. After the extended communication protocol is configured, a unique communication station number is assigned to each drive motor in the drive motor group. Establish a communication addressing benchmark that characterizes the mapping relationship between each drive motor and the communication station number; The target control parameters of each drive motor are obtained by parsing the overall machine control command. Based on the target control parameters of each drive motor, an initial motor speed command set is generated, which includes the angular velocity of the transplanter and the linear velocity of the walking motor and corresponds to the communication addressing reference. The pre-collected encoder feedback parameters, current loop sampling parameters, and adjacent drive motor state parameters are input into the Kalman filter for fusion calculation to obtain the actual operating state of each drive motor. Based on the actual operating state of each drive motor, the initial motor speed command set is corrected to obtain the corrected motor speed command set. A global time reference criterion is established using a pulse coupling synchronization algorithm. Under the constraints of the global time reference criterion, the coupling strength between each drive motor is adjusted according to the corrected motor speed command set, and a phase-aligned motor group synchronization control command is output.

[0009] Preferably, the step of establishing a global time reference criterion using a pulse coupling synchronization algorithm, and adjusting the coupling strength between each drive motor according to the corrected motor speed command set under the constraints of the global time reference criterion, and outputting phase-aligned motor group synchronization control commands includes: For each drive motor in the drive motor group, a corresponding pulse coupling oscillator is constructed, and a unified counting period and reference beat are set for each pulse coupling oscillator. A global time reference criterion for synchronization of each drive motor is established using the reference beat generator as a carrier. Based on the global time base criterion and the initial state parameters of the pulse-coupled oscillators, a preset linear phase function is used to periodically accumulate and update the phase state of each pulse-coupled oscillator to obtain the updated phase state of each pulse-coupled oscillator. The update phase state of each pulse-coupled oscillator is monitored in real time. When the update phase state of any pulse-coupled oscillator meets the preset phase excitation threshold, the pulse-coupled oscillator is determined to enter the excitation state. After the pulse-coupled oscillator completes the excitation action, the corresponding excitation state signal and coupling strength parameter are output. The excitation state signal and the corresponding coupling strength parameter are broadcast to the pulse-coupled oscillator in the non-excitation state, and the phase state of the pulse-coupled oscillator in the non-excitation state is updated by phase increment using the phase formula. The coupling strength between each pulse-coupled oscillator is dynamically adjusted based on the phase state of the pulse-coupled oscillator updated by the phase increment and in conjunction with the corrected motor speed command set. Based on the coupling strength between each pulse-coupled oscillator, a shielding period verification is performed on each pulse-coupled oscillator, and a phase-aligned synchronous control command for the motor set is generated based on the shielding period verification result.

[0010] Preferably, the step of performing a shielding period check on each pulse-coupled oscillator based on the coupling strength between them, and generating a phase-aligned motor synchronization control command based on the shielding period check result, includes: After the pulse-coupled oscillator completes the excitation action, a shielding period of preset duration is initiated based on the coupling strength between each pulse-coupled oscillator. During the shielding period, each pulse-coupled oscillator maintains its own periodic cumulative update mechanism, shields itself from receiving the coupling strength parameters of other pulse-coupled oscillators, and temporarily does not perform phase increment updates; after the shielding period ends, phase increment updates are resumed, and the phase naturally accumulated during the shielding period is used as the effective phase parameter output. Based on the effective phase parameters of the output, compare the real-time phase deviation of all pulse-coupled oscillators; When the phase deviation is within the synchronization convergence threshold range, it is determined that the drive motor group has achieved phase alignment of multiple drive motors, and the motor group synchronization control command is generated based on the aligned unified phase and the corrected motor speed command set; otherwise, the phase accumulation update and synchronization convergence process is re-executed.

[0011] Preferably, the transplanting execution unit includes a driving status perception module, a collaborative instruction generation module, and a seedling placement collaborative drive module; Among them, the driving status perception module is used to analyze and extract the real-time driving speed and driving phase reference of the whole vehicle based on the driving status synchronization reference signal; The collaborative instruction generation module is used to convert the real-time driving speed and the walking phase reference into collaborative action timing instructions for the seedling dispensing mechanism and the duckbill mechanism that are linked to the driving state, based on the driving speed-planting frequency matching rule. The seedling feeding collaborative drive module is used to drive the seedling feeding mechanism and the duckbill mechanism to move sequentially according to the collaborative action timing instructions through the terrain adaptive contouring algorithm. During this process, action execution and feedback data are collected in real time. The transplanting execution parameter calibration and output module is used to sequentially perform filtering and noise reduction, outlier removal and dimensional calibration operations on the action execution and feedback data, generate standardized transplanting action parameters and output them to the ridge planting posture optimization unit.

[0012] Preferably, the seedling dispensing collaborative drive module, when driving the seedling dispensing mechanism and the duckbill mechanism sequentially according to the collaborative action timing instructions using a terrain adaptive contouring algorithm, includes: Real-time data collection of machine acceleration and planting resistance time-series data is used to construct a fused dataset. Nonlinear temporal characteristics representing ridge topographic undulation and soil planting resistance are extracted from the fused dataset, and phase space reconstruction is performed on the fused dataset based on the nonlinear temporal characteristics to obtain high-dimensional features representing topographic changes. The high-dimensional features representing topographic changes are input into a pre-built weighted local prediction model. The weighted local prediction model is used to predict the changing trends of ridge topographic undulation and soil planting resistance in future time periods and output ultra-short-term topographic prediction results. The ultra-short-term terrain prediction results are nonlinearly mapped to advance correction values ​​using a Kalman filter. Based on the advance correction values ​​and the timing instructions of the coordinated actions, the duckbill mechanism is driven to complete the adaptive contour insertion action. The duckbill mechanism is then driven to cooperate in completing the adaptive contour insertion action.

[0013] Preferably, the step of extracting nonlinear temporal characteristics representing ridge topographic undulation and soil planting resistance from the fused dataset, and performing phase space reconstruction on the fused dataset based on the nonlinear temporal characteristics to obtain high-dimensional features representing topographic changes includes: Extract the corresponding length of the whole machine driving acceleration time series and planting resistance time series from the fused dataset according to the target sliding window; Calculate the maximum Lyapunov exponent of the two time series segments. Only when the maximum Lyapunov exponent of both time series segments is greater than zero is the two time series segments determined to have nonlinear time series characteristics and combined into an effective fused time series set. If the determination condition is not met, return to adjust the length of the target sliding window. Using the effectively fused time series set as input, the dimensionless statistics are calculated through correlation integrals; the delay time of the two time series segments is determined by the first minimum point of the dimensionless statistics curve, the optimal time window is determined by the global minimum point of the dimensionless statistics curve, and the embedding dimension of the two time series segments is calculated according to the correlation formula between the optimal time window and the embedding dimension. Using time delay and embedding dimension as input conditions, phase space reconstruction is performed on the effectively fused time series based on the temporal high-dimensional embedding criterion, mapping one-dimensional time series data to a high-dimensional phase space, and mining high-dimensional features that characterize the variation of ridge topography and soil planting resistance.

[0014] Preferably, the step of using delay time and embedding dimension as input conditions, and performing phase space reconstruction on the effectively fused time series set based on the temporal high-dimensional embedding criterion to map one-dimensional time series data to a high-dimensional phase space, and mining high-dimensional features characterizing the variation law of ridge topography undulation and soil planting resistance includes: Based on the delay time and embedding dimension of the two sets of time segments, high-dimensional embedding vectors of the two sets of time segments are constructed respectively. Based on the global unified time reference of the coordinated action timing instructions, the timestamp vectors of the two sets of high-dimensional embedded vectors are horizontally synchronously spliced ​​to generate a reconstructed phase space matrix of multivariate fusion. Based on the reconstruction of the phase space matrix by multivariate fusion, the evolution trajectory features of the state vector in the phase space are extracted, redundant noise components in the evolution trajectory features are removed, and only the high-dimensional feature set of terrain changes that can completely characterize the changes in ridge topography and soil planting resistance are retained. The high-dimensional feature set of terrain changes is sequentially subjected to false nearest neighbor verification and feature consistency verification. After the verification is passed, the high-dimensional features representing the terrain changes are obtained.

[0015] Preferably, the step of inputting high-dimensional features characterizing terrain changes into a pre-constructed weighted local prediction model, and predicting the changing trends of ridge topography undulation and soil planting resistance in future time periods and outputting ultra-short-term terrain prediction results through the weighted local prediction model includes: A weighted local prediction model is constructed based on the Euclidean distance between neighboring phase points in high-dimensional phase space and the local linear fitting rule. The original fusion dataset containing the whole machine driving acceleration data and planting resistance time series data is divided into a training set and a validation set. The training set is input into the weighted local prediction model to perform iterative training. After each round of training, the weighted local prediction model is validated through the validation set. The trained weighted local prediction model is obtained when the maximum number of iterations is met. The high-dimensional features characterizing topographic changes are input into the weighted local prediction model. The weighted local prediction model locates the high-dimensional phase point corresponding to the current high-dimensional feature in the reconstructed high-dimensional phase space and selects the neighboring phase point with the closest dynamic characteristics to the high-dimensional phase point. The weighted local prediction model uses the reciprocal of the Euclidean distance between the current high-dimensional phase point and its neighboring phase points as the weighting coefficients. Based on the preset local linear fitting rules, it completes the fitting calculation and obtains the evolution trajectory of the high-dimensional phase point within a preset time period in the future. The evolution trajectory of high-dimensional phase points is inversely mapped to the changing trend of ridge topography undulation and soil planting resistance, and the ultra-short-term topographic prediction results are output for duckbill mechanism and seedling placement mechanism.

[0016] The beneficial effects of this invention are as follows: 1. This invention assigns a unique communication station number to each drive motor and establishes a communication addressing benchmark by combining an extended communication protocol with a dual verification mechanism, ensuring accurate and reliable motor communication addressing. It then parses the overall machine control commands to generate an initial motor speed command set. Using a Kalman filter to fuse encoder, current loop, and adjacent motor state parameters, it accurately identifies the actual operating state of the motor and corrects the speed command set. Simultaneously, it configures an independent pulse coupling oscillator for each drive motor through a pulse coupling synchronization algorithm. Based on a unified benchmark beat, it constructs a global time benchmark criterion. Combined with phase periodic cumulative update, phase excitation judgment, broadcast phase increment correction, dynamic coupling strength adjustment, and shielding period verification and phase deviation convergence judgment mechanisms, it achieves precise phase alignment and synchronous collaborative control of multiple drive motors. This effectively suppresses motor synchronization errors under field speed fluctuations, load disturbances, and steering conditions, avoiding problems of disordered transplanting rhythm. It provides a reliable synchronous control foundation for precise timing matching between the seedling feeding mechanism and the duckbill mechanism, and for stable and high-quality transplanting operations.

[0017] 2. This invention extracts time-series segments through a sliding window and uses the maximum Lyapunov exponent to identify nonlinear chaotic characteristics. It accurately calculates the delay time, optimal time window, and embedding dimension by combining correlation integrals and dimensionless statistics. Based on the Tukens embedding theorem, it completes the reconstruction of multivariate phase space and extracts reliable high-dimensional terrain features after verification. Then, it constructs and trains a weighted local prediction model by Euclidean distance of neighboring phase points and local linear fitting, realizing ultra-short-term prediction of the ridge topography undulation and soil planting resistance change trend. Based on this, the seedling placement mechanism and the duckbill mechanism are driven by coordinated time-series commands to complete adaptive contour planting, effectively realizing the precise linkage and matching of driving status and transplanting action. It has the ability to perceive and adaptively compensate for changes in ridge topography and soil resistance, thereby suppressing problems such as uneven plant spacing, missing seedlings, overlapping seedlings, pinched seedlings, and planting depth deviation. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram illustrating the principle operation of an intelligent planter for tobacco seedlings according to an embodiment of the present invention. Figure 2 This is a schematic diagram of an intelligent planter for tobacco seedlings according to an embodiment of the present invention.

[0020] In the picture: 1. Whole machine control unit; 2. Motor unit coordination control unit; 3. Walking drive and steering unit; 4. Transplanting execution unit; 5. Ridge planting posture optimization unit. Detailed Implementation

[0021] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0022] According to an embodiment of the present invention, an intelligent planter for tobacco seedlings is provided.

[0023] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figures 1-2 As shown, according to an embodiment of the present invention, an intelligent planter for tobacco seedlings includes five parts: a walking power assembly, a transplanting execution assembly, a frame adjustment assembly, a human-machine interaction assembly, and an electronic control assembly.

[0024] The walking power assembly includes two geared motors of the same specification, namely the left walking motor and the right walking motor, which serve as the walking power source for the whole machine. The left and right steering of the whole machine is achieved by the differential rotation of the left and right walking motors. The transplanting execution assembly is equipped with an independent transplanting drive motor, which is used to drive the seedling feeding mechanism and the duckbill planting mechanism. The three drive mechanisms form a drive motor group.

[0025] The transplanting assembly includes a transplanting drive motor, a seedling feeding mechanism, and a duckbill planting mechanism. The seedling feeding mechanism and the duckbill planting mechanism achieve synchronous matching of their action sequences through a gearbox and chain transmission mechanism. The transplanting drive motor provides driving power for the seedling feeding mechanism and the duckbill planting mechanism. The frame adjustment assembly includes balance guide wheels, a planting depth adjustment component, a ridge width adjustment component, and a plant spacing adjustment component. The balance guide wheels are arranged in an inward V-shape, moving along both sides of the ridge during operation to guide and limit the overall movement of the machine. The planting depth adjustment component is used to adjust the relative height and parallelism between the planter and the ridge surface, thereby adjusting the planting depth and verticality of the seedlings. The ridge width adjustment component uses a bolt locking structure to adjust the lateral width, adapting to different ridge widths. The plant spacing adjustment component uses a core algorithm to match the travel speed with the planting frequency, achieving precise adjustment of the plant spacing for seedling planting. The human-machine interface assembly includes a speed control throttle, a forward / reverse switch, a mode switch, and an emergency stop button. The speed control throttle is used to adjust the overall speed in manual mode, the forward / reverse switch is used to switch the overall direction of travel, the mode switch is used to switch between automatic and manual modes, and the emergency stop button is used for emergency stop control of the overall machine. The electronic control assembly includes a motor driver, an electromagnetic brake, and a main controller. The electromagnetic brake is configured one-to-one with the left and right travel motors and adopts a power-off lock-up braking mode. After power failure, the corresponding drive motor is directly locked to realize the machine's parking on slopes and emergency braking.

[0026] The machine has two working modes: one is automatic mode, which is an automatic seedling planting mode. After the plant spacing parameters are preset, the machine automatically completes the entire process of driving along the ridge, seedling placement, and contour planting. The other is manual mode, which is a pure driving mode. The machine's driving speed is controlled by the speed adjustment throttle, and the driving direction is controlled by the forward and reverse switch. It is used for moving the machine to different locations and adjusting the working conditions.

[0027] The aforementioned walking power assembly, transplanting execution assembly, frame adjustment assembly, human-machine interface assembly, and electronic control assembly constitute the physical foundation for the mechanical load-bearing, power execution, hardware interaction, and safety protection of the intelligent tobacco seedling planter, providing complete hardware support for the entire machine's field transplanting operations. To achieve high-precision coordinated operation of each hardware assembly, adaptive adaptation to complex field conditions, and stable execution of the fully automated transplanting operation, this invention is equipped with a hierarchical intelligent control architecture deeply coupled with the aforementioned hardware assemblies. This architecture is divided into five core control units according to functional boundaries, with each unit forming a one-to-one driving, feedback, and control linkage relationship with the hardware assembly.

[0028] The overall machine control unit, as the top-level control center, connects to the main controller of the human-machine interface assembly and the electronic control assembly, coordinating the switching of machine modes, safety interlocks, and command issuance. The motor unit coordination control unit, relying on the communication and drive module of the electronic control assembly, connects to the drive motor units of the walking power assembly and the transplanting execution assembly, realizing communication addressing, synchronous control, and status management of multiple motors. The walking drive and steering unit directly drives the walking power assembly, executing the machine's driving and differential steering actions. The transplanting execution unit connects to the transplanting execution assembly, completing the timing coordination and precise execution of seedling placement and planting actions. The ridge planting posture optimization unit connects to the frame adjustment assembly and the transplanting execution assembly, realizing adaptive optimization and closed-loop adjustment of planting posture based on operation feedback data. The specific functions and implementation methods of each control unit are described in detail below: The whole machine control unit 1 is used to integrate the pre-acquired whole machine operating condition control signals and safety interlock control signals into whole machine control commands, and send the whole machine control commands to the motor set cooperative control unit 2.

[0029] Specifically, the whole machine control unit 1 is used to collect in real time the whole machine operating condition control signals such as mode switching, speed adjustment, and direction switching from the human-machine interface assembly. At the same time, it collects safety interlock control signals such as emergency stop signal, limit signal, motor fault signal, transplanting mechanism positioning signal, and electromagnetic brake status. The key safety signals adopt a dual interlock protection mechanism of hardware AND gate and software verification. The transplanting mechanism positioning signal and the walking enable signal are logically connected through hardware AND gate circuit. The emergency stop button signal directly cuts off the enable circuit of all motor drivers. After the limit switch is triggered, the software automatically starts the reverse deceleration curve to achieve a stop buffer to avoid mechanical impact. The whole machine control unit 1 filters, logically judges and verifies the validity of the above operating condition control signals and safety interlock signals, integrates them into whole machine control commands according to preset control rules, and sends the whole machine control commands to the motor group collaborative control unit 2. At the same time, it receives the updated operating parameters returned by the ridge planting posture optimization unit 5, completes parameter refresh and generates a new round of optimized whole machine control commands, realizing closed-loop scheduling and safety management of the whole system.

[0030] The motor unit coordination control unit 2 is used to allocate communication station numbers to the drive motor unit based on the extended communication protocol to establish a motor communication addressing reference. The global time reference criterion established by the pulse coupling synchronization algorithm is used as a constraint condition, and the drive motor unit synchronization control command is generated according to the whole machine control command and the communication addressing reference.

[0031] Among them, the motor set cooperative control unit 2, when allocating communication station numbers to the drive motor set based on the extended communication protocol to establish a motor communication addressing reference, and using the global time reference criterion established by the pulse coupling synchronization algorithm as a constraint condition, generates synchronous control commands for the drive motor set based on the overall machine control commands and the communication addressing reference, includes: The extended communication protocol is configured using a dual verification mechanism combining parity check and cyclic redundancy check. After the extended communication protocol is configured, a unique communication station number is assigned to each drive motor in the drive motor group.

[0032] Specifically, the extended communication protocol is an event-triggered enhanced Modbus-RTU protocol, which extends the standard Modbus frame structure with multiple layers. The specific configuration is as follows: Physical layer and data link layer configuration: RS-485 differential transmission circuit is adopted, and a dual verification mechanism combining hardware parity check and software CRC-16 cyclic redundancy check is configured; the bus communication rate adopts an adaptive adjustment mode, with a default communication rate of 115200bps. When three consecutive data packets are detected with CRC check errors, the speed is automatically reduced to 9600bps and bit error rate monitoring is started. The default communication rate is restored after the bit error rate is lower than the preset threshold. Motor station number allocation rules: Each drive motor in the drive motor group is assigned an independent and unique communication station number. The station number of the transplant drive motor is configured as 0x01, the station number of the left travel motor is configured as 0x02, and the station number of the right travel motor is configured as 0x03. Broadcast Collaborative Writing Mechanism: A specially designed broadcast collaborative writing function code 0x1A is included. The frame structure contains a 1-byte fixed broadcast address (0xFF), a 1-byte 0x1A function code, a 1-byte data length field, 2 bytes of transplanter speed data, 2 bytes of left walking speed data, 2 bytes of right walking speed data, and a 2-byte dynamic CRC16 check field based on timestamps. The transplanter speed is a 16-bit signed integer with a precision of 0.1 RPM, and the walking speed is a 16-bit signed integer with a precision of 0.01 m / s. This mechanism can reduce the collaborative command transmission delay from 15 ms in the traditional polling mode to less than 3 ms, and control the synchronization error to less than 50 μs. Anti-interference and fault tolerance mechanisms: The protocol layer implements triple fault tolerance protection; the physical layer adds a preamble 0xAA55 to each frame of data as a frame start marker; the data link layer adopts a sliding window protocol, supporting up to 3 consecutive retransmissions; and the application layer contains a unique sequence number for each data packet, supporting out-of-order reassembly and packet loss detection.

[0033] Establish a communication addressing benchmark that characterizes the mapping relationship between each drive motor and the communication station number.

[0034] Specifically, after completing the dual verification configuration, rate adaptive adjustment, anti-interference and triple fault tolerance mechanism configuration of the extended communication protocol, unique communication station numbers 0x01, 0x02 and 0x03 are assigned to the transplanting drive motor, left travel motor and right travel motor in the drive motor group according to the preset motor station number allocation rules. The device number, physical address, control function and data type of each drive motor are bound to the corresponding communication station number, forming a fixed mapping relationship between motor identifier and communication station number. Based on this mapping relationship, a communication addressing benchmark that can be globally identified and called is established, enabling the motor group collaborative control unit to quickly locate the target drive motor, accurately send control commands and receive status feedback according to the communication addressing benchmark, providing a stable and reliable communication addressing foundation for the generation and issuance of subsequent multi-motor synchronous control commands.

[0035] The target control parameters of each drive motor are obtained by parsing the overall machine control command. Based on the target control parameters of each drive motor, an initial motor speed command set is generated, which includes the angular velocity of the transplanter and the linear velocity of the walking motor and corresponds to the communication addressing reference. The pre-collected encoder feedback parameters, current loop sampling parameters, and adjacent drive motor state parameters are input into a Kalman filter for fusion calculation to obtain the actual operating state of each drive motor. Based on the actual operating state of each drive motor, the initial motor speed command set is corrected to obtain the corrected motor speed command set, which specifically includes: The unscented Kalman filter is initialized by using the target speed, initial phase, and preset no-load torque of each drive motor in the initial motor speed command set as the initial state estimate, and the corresponding state estimate error covariance matrix is ​​set to complete the filter initialization. With a fixed sampling period aligned with the global time base, the encoder feedback pulse count of each drive motor, the armature current value sampled by the current loop, and the real-time speed difference between adjacent drive motors are synchronously collected as observation input data. Based on the current state estimate, a set of representative state sampling points around the estimate is generated. Each sampling point corresponds to a set of possible combinations of motor speed, phase, and load torque. Each state sampling point is then input into a pre-established motor dynamics model to perform state prediction, thereby obtaining the predicted state value of each sampling point at the next moment. At the same time, all predicted state values ​​are weighted and fused to obtain the system state prediction value at the next moment. The predicted state value of each state sampling point is mapped to the observation space to obtain the corresponding predicted observation value. Similarly, all predicted observation values ​​are weighted and fused to obtain the system observation prediction value. Then, the actual collected observation input data is compared with the system observation prediction value to calculate the observation residual. Based on the observation residual and the preset observation noise covariance and state noise covariance, the system state prediction value is corrected and updated to obtain the optimal estimate of the actual speed, actual phase and actual load torque of each drive motor at the current moment, that is, the actual operating state of each drive motor. The actual operating status of each drive motor is compared with the target speed and target phase in the initial motor speed command set. The speed tracking error and phase deviation are calculated. Based on the error and deviation, a proportional-integral control strategy is used to dynamically correct the initial motor speed command set, eliminating the influence of load disturbance and tracking error, and obtaining the corrected motor speed command set, which provides accurate input commands for the subsequent pulse coupling synchronization algorithm.

[0036] The unscented Kalman filter is an optimal state estimation algorithm for nonlinear systems. It is an improved version of the traditional Kalman filter. It addresses the accuracy loss and divergence problems caused by the nonlinear linearization approximation of the extended Kalman filter. It generates a finite number of sampling points through the deterministic sampling strategy of the unscented transform to completely approximate the probability distribution of the nonlinear system state. It does not require the calculation of a complex Jacobian matrix and can effectively fuse multi-source sensor data with noise. It is an existing technology and will not be elaborated on further here.

[0037] A global time reference criterion is established using a pulse coupling synchronization algorithm. Under the constraints of the global time reference criterion, the coupling strength between each drive motor is adjusted according to the corrected motor speed command set, and a phase-aligned motor group synchronization control command is output.

[0038] Specifically, a global time reference criterion is established using a pulse coupling synchronization algorithm. Under the constraints of the global time reference criterion, the coupling strength between each drive motor is adjusted according to the corrected motor speed command set, and the output phase-aligned motor group synchronization control commands include: For each drive motor in the drive motor group, a corresponding pulse coupling oscillator is constructed, and a unified counting period and reference beat are set for each pulse coupling oscillator. A global time reference criterion for synchronization of each drive motor is established using the reference beat generator as a carrier.

[0039] Based on the global time base criterion and the initial state parameters of the pulse-coupled oscillators, a preset linear phase function is used to periodically accumulate and update the phase state of each pulse-coupled oscillator to obtain the updated phase state of each pulse-coupled oscillator. The update phase state of each pulse-coupled oscillator is monitored in real time. When the update phase state of any pulse-coupled oscillator meets the preset phase excitation threshold, the pulse-coupled oscillator is determined to enter the excitation state. After the pulse-coupled oscillator completes the excitation action, the corresponding excitation state signal and coupling strength parameter are output. The excitation state signal and the corresponding coupling strength parameter are broadcast to the pulse-coupled oscillator in the non-excitation state, and the phase state of the pulse-coupled oscillator in the non-excitation state is updated by phase increment using the phase formula. The coupling strength between each pulse-coupled oscillator is dynamically adjusted based on the phase state of the pulse-coupled oscillator updated by the phase increment and in conjunction with the corrected motor speed command set. Based on the coupling strength between each pulse-coupled oscillator, a shielding period verification is performed on each pulse-coupled oscillator, and a phase-aligned synchronous control command for the motor set is generated based on the shielding period verification result.

[0040] The following section further explains how to establish a global time reference criterion using a pulse coupling synchronization algorithm, and how, under the constraints of the global time reference criterion, the coupling strength between each drive motor is adjusted according to the corrected motor speed command set to output phase-aligned motor group synchronization control commands.

[0041] Independent pulse-coupled oscillators (PCOs) are constructed for the transfer drive motor, left travel motor, and right travel motor in the drive motor group. A unified counting period and reference beat are set for each pulse-coupled oscillator. The basic period T0 of the reference beat generator is set to 200ms, which is a parameter that can be adjusted according to the working conditions. The counting period of the reference beat generator is completely aligned with the broadcast collaborative write frame transmission period of the extended Modbus-RTU protocol mentioned above. The counting start point of all pulse-coupled oscillators is synchronized through the broadcast frame. Based on this, a global time reference criterion for complete synchronization of all drive motors is established. The core formula for phase update of each pulse-coupled oscillator is: (Base cycle, adjustable); In the formula, Indicates the first Phase angle of each motor; Indicates the natural frequency; Indicates coupling strength K represents the coupling function; Indicates the first j Phase angle and natural frequency of each motor The speed of the motor is linearly proportional to the target speed in the corrected motor speed command set, where K is the coupling function, in standard sinusoidal form, and its specific expression is as follows: , where is the phase difference between the two oscillators; Based on the global time base criterion and the initial state parameters of the pulse-coupled oscillator with an initial phase of 0, a preset linear phase function is adopted. The phase state of each pulse-coupled oscillator is periodically accumulated and updated, where Δ T The minimum counting step size for the global time reference is consistent with the sampling period of the motor encoder to obtain the updated phase state of each pulse-coupled oscillator; the updated phase state of each pulse-coupled oscillator is monitored in real time, and the preset phase excitation threshold is set to 2π. When the updated phase state of any pulse-coupled oscillator accumulates to 2π, it is determined that the pulse-coupled oscillator has entered the excitation state. After the oscillator completes the excitation action, the phase is automatically reset to 0, and the corresponding excitation state signal and the current coupling strength parameter are broadcast to the bus. The excitation state signal and the corresponding coupling strength parameter are broadcast to all non-excited pulse-coupled oscillators, and the phase state of the non-excited pulse-coupled oscillators is updated by phase increment using the phase increment update formula. Based on the phase state of the pulse-coupled oscillators updated by the phase increment, and combined with the corrected motor speed command set, the coupling strength between each pulse-coupled oscillator is dynamically adjusted. When the phase deviation between motors is greater than a preset threshold, the coupling strength is increased to accelerate synchronization convergence; when the phase deviation is less than the preset threshold, the coupling strength is decreased to reduce system disturbance. Based on the coupling strength between each pulse-coupled oscillator, a shielding period verification is performed on each pulse-coupled oscillator, and a phase-aligned motor synchronization control command is generated based on the shielding period verification result. Specifically, after the pulse-coupled oscillator completes its excitation action, a shielding period of a preset duration is initiated based on the coupling strength. The shielding period duration is set to... T0 is the basic period of the reference beat. During the shielding period, each pulse-coupled oscillator maintains its own periodic cumulative update mechanism, shields the receiving of the coupling strength parameters of other pulse-coupled oscillators, and temporarily does not perform phase increment update. After the shielding period ends, phase increment update is resumed, and the phase naturally accumulated during the shielding period is used as the effective phase parameter output. Based on the effective phase parameters of the output, the real-time phase deviation of all pulse-coupled oscillators is compared in real time. When the phase deviation of all motor corresponding oscillators is within the preset synchronization convergence threshold range, i.e., the phase deviation is <0.1π, it is determined that the drive motor group has achieved phase alignment of multiple drive motors. Based on the aligned unified phase and the corrected motor speed command set, the motor group synchronization control command is generated and sent to each drive motor through the broadcast collaborative writing function code of the extended Modbus-RTU protocol. Otherwise, the phase accumulation update and synchronization convergence process is re-executed.

[0042] The process includes: performing a shielding period check on each pulse-coupled oscillator based on the coupling strength between them; and generating phase-aligned motor synchronization control commands based on the shielding period check results. After the pulse-coupled oscillator completes the excitation action, a shielding period of preset duration is initiated based on the coupling strength between each pulse-coupled oscillator. During the shielding period, each pulse-coupled oscillator maintains its own periodic cumulative update mechanism, shields itself from receiving the coupling strength parameters of other pulse-coupled oscillators, and temporarily does not perform phase increment updates; after the shielding period ends, phase increment updates are resumed, and the phase naturally accumulated during the shielding period is used as the effective phase parameter output. Based on the effective phase parameters of the output, compare the real-time phase deviation of all pulse-coupled oscillators; When the phase deviation is within the synchronization convergence threshold range, it is determined that the drive motor group has achieved phase alignment of multiple drive motors, and the motor group synchronization control command is generated based on the aligned unified phase and the corrected motor speed command set; otherwise, the phase accumulation update and synchronization convergence process is re-executed.

[0043] The walking drive and steering unit 3 is used to use the drive motor set as the walking power source, coordinate the operation status between each drive motor based on the synchronous control command of the drive motor set to perform driving and steering actions, and synchronously output the driving status reference signal to the transplanting execution unit 4.

[0044] Specifically, the speed closed-loop control of the walking motor adopts a composite control architecture of fuzzy PID and iterative learning. The first layer is a fuzzy PID controller, which takes speed error and error change rate as input and PID parameter correction as output. It completes parameter adaptive adjustment through 49 7×7 fuzzy rules and uses the centroid method to complete defuzzification. The second layer is an iterative learning controller, which realizes iterative compensation of speed error under repetitive operation conditions based on the Toplitz matrix of the system transfer function, thereby improving speed tracking accuracy.

[0045] The entire machine monitors the motor current spectrum in real time, performs a 1024-point FFT transformation on the current signal in the 0.5Hz-100Hz frequency band, calculates the power spectral density (PSD), identifies frequency components whose peak values ​​exceed the baseline by 3σ, and determines mechanical resonance when the peak frequency is within the range of 0.8-1.2 times the mechanical natural frequency. It then activates an adaptive notch filter group to suppress the resonance. The pole radius of the notch filter is adaptively adjusted according to the amplitude of the resonance peak value, while the zero radius is fixed. This eliminates the resonance component without affecting the dynamic response of the system.

[0046] The transplanting execution unit 4 is used to generate coordinated action timing instructions that match the seedling feeding mechanism, the duckbill mechanism and the driving status reference signal according to the preset driving speed-planting frequency matching rules. It drives the seedling feeding mechanism to perform seedling feeding action and drives the duckbill mechanism to cooperate in completing adaptive contour planting action according to the coordinated action timing instructions. In the process of performing seedling feeding and planting actions, it acquires action execution and feedback data in real time and transmits them to the ridge planting posture optimization unit 5.

[0047] Among them, the transplanting execution unit 4 includes a driving status perception module, a collaborative instruction generation module, and a seedling placement collaborative drive module; The driving status perception module is used to analyze and extract the real-time driving speed and driving phase reference of the vehicle based on the driving status synchronization reference signal.

[0048] It should be noted that the driving status perception module acquires and decodes the driving status synchronization reference signal in real time to obtain the encoder pulse count, speed feedback value and global time reference phase signal output by the pulse coupling synchronization algorithm of the left and right driving motors. Based on the motor reduction ratio, wheel rolling radius and encoder line count, the speed is converted to obtain the real-time driving speed of the whole vehicle. At the same time, using the global time reference as a synchronization reference, the motor running phase is phase-locked tracking and standardized and normalized to extract the driving phase reference that is strictly synchronized with the driving rhythm of the whole vehicle and can be used for the coordinated matching of seedling placement and planting actions.

[0049] The collaborative instruction generation module is used to convert the real-time driving speed and the walking phase reference based on the driving speed-planting frequency matching rule to generate collaborative action timing instructions for the seedling placement mechanism and the duckbill mechanism that are linked to the driving state.

[0050] Specifically, the speed-planting frequency matching rule refers to using a preset seedling planting spacing as a constant target, and dynamically adjusting the planting frequency according to the real-time speed of the entire vehicle to maintain a linear proportional constraint relationship between the speed and the planting frequency. At the same time, using the travel phase reference as a synchronization reference, it ensures that the timing of the seedling feeding mechanism and the duckbill mechanism is strictly aligned with the overall machine travel phase. Specifically, with the plant spacing L as a fixed set value, the real-time speed v is converted into the corresponding planting frequency f using the formula f=v / L. Then, using the period of the travel phase reference as a time reference, the seedling feeding triggering sequence and the duckbill opening and closing sequence are generated according to the planting frequency and synchronized with the travel phase. This ensures that the entire machine completes one seedling feeding and planting action every time it travels a fixed plant spacing distance, so that the plant spacing remains constant when the travel speed changes. At the same time, the timing of the seedling feeding mechanism and the duckbill mechanism are linked in real time with the travel status.

[0051] The seedling feeding collaborative drive module is used to drive the seedling feeding mechanism and the duckbill mechanism to move sequentially according to the collaborative action timing instructions through the terrain adaptive contouring algorithm. During this process, action execution and feedback data are collected in real time.

[0052] Specifically, the terrain-adaptive contouring algorithm is based on real-time collected machine acceleration and planting resistance time-series data. Through nonlinear time-series characteristic extraction, phase space reconstruction, weighted local prediction, and Kalman filter mapping, it achieves ultra-short-term look-ahead prediction of ridge topography undulations and soil planting resistance changes. The prediction results are transformed into advance corrections for the height, angle, insertion depth, and seedling placement timing of the duckbill mechanism. This allows the duckbill mechanism to adaptively adjust in real time according to the ridge topography undulations, always maintaining a stable planting depth and seedling verticality. This algorithm achieves coordinated seedling placement... When the drive module drives the seedling feeding mechanism and the duckbill mechanism to move according to the coordinated action timing instructions, it simultaneously collects action execution and feedback data such as acceleration, planting resistance, motor current, mechanism displacement, planting depth deviation, and phase synchronization error. The above data is transmitted in real time to the ridge planting posture optimization unit 5 to generate full-domain adjustment instructions for guiding driving, height parallelism adjustment, ridge width adjustment, and plant spacing correction. Ultimately, it achieves an adaptive transplanting effect with consistent planting depth, stable seedling uprightness, accurate plant spacing, smooth operation, and no mechanical impact under complex field terrain.

[0053] The seedling dispensing collaborative drive module, when using a terrain-adaptive contouring algorithm and sequentially driving the seedling dispensing mechanism and the duckbill mechanism according to the collaborative action timing instructions, includes: Real-time data collection of the machine's acceleration and planting resistance time series data is used to construct a fused dataset.

[0054] Specifically, the longitudinal, lateral, and vertical acceleration time-series signals of the whole machine during the driving process are collected in real time. The soil resistance time-series signals during the duckbill insertion and planting process are collected in real time by the tension or pressure sensor installed at the drive end of the duckbill planting mechanism. At the same time, the encoder position signals, current sampling signals, and pulse coupling synchronization phase signals of the left travel motor, right travel motor, and transplant drive motor are collected synchronously. The above-mentioned multiple signals are timestamped and synchronously sampled with a globally unified time reference. The acceleration time-series data, planting resistance time-series data, and motor operation status data are integrated into a standardized fusion dataset according to the time series.

[0055] Nonlinear temporal characteristics representing ridge topographic undulation and soil planting resistance are extracted from the fused dataset, and phase space reconstruction is performed on the fused dataset based on the nonlinear temporal characteristics to obtain high-dimensional features representing topographic changes.

[0056] Specifically, nonlinear temporal characteristics representing ridge topographic undulation and soil planting resistance are extracted from the fused dataset, and phase space reconstruction is performed on the fused dataset based on the nonlinear temporal characteristics to obtain high-dimensional features representing topographic changes, including: Extract the corresponding length of the whole machine driving acceleration time series and planting resistance time series from the fused dataset according to the target sliding window; Calculate the maximum Lyapunov exponent for the two time series segments. Only when the maximum Lyapunov exponent for both time series segments is greater than zero is the two time series segments determined to have nonlinear time series characteristics and combined into an effective fused time series set. If the determination condition is not met, return to adjust the length of the target sliding window.

[0057] It should be noted that the Wolf method is used to calculate the maximum Lyapunov exponent of the two time series segments separately. The maximum Lyapunov exponent of the two time series segments is compared with the preset threshold of 0. Only when the maximum Lyapunov exponent of both time series segments is greater than 0 is it determined that the fused time series dataset has nonlinear time series characteristics that are strongly correlated with the topography and soil characteristics, and the valid fused time series data verified by the characteristics is output. If the judgment condition is not met, the length of the target sliding window is adjusted, and the process returns to the beginning of this step to re-extract the time series segments and perform calculation and verification.

[0058] Specifically, according to the target sliding window, time-series segments of whole-machine driving acceleration and planting resistance of corresponding length are extracted from the fused dataset. The initial length of the target sliding window is set to the number of sampling points corresponding to two seedling planting cycles, and the sliding step size is a single sampling cycle. The two extracted time-series segments maintain perfect timestamp alignment and have completely consistent sampling length and sampling frequency. The improved Wolf method adapted to short time-series data is used to calculate the maximum Lyapunov exponent of the two time-series segments respectively. The specific calculation process is as follows: Centering and normalization preprocessing are performed on the two time segments respectively to eliminate dimensional differences and DC component interference. To meet the phase space fundamental requirements for the calculation of the maximum Lyapunov exponent, a common short-time chaotic initial screening fixed empirical value is adopted, and the initial embedding dimension m=6 and the initial delay time are set. Complete the phase space pre-reconstruction and construct a pre-reconstructed high-dimensional embedding vector for this exponent calculation only: Where X(t) is the time series data value at time t, and the range of t is... N is the total number of sampling points in the time segment. This pre-reconstruction result is only used for the calculation of the maximum Lyapunov exponent and does not participate in any subsequent feature extraction and model prediction. In the pre-reconstructed phase space, locate the nearest neighbor phase point of each initial phase point. Limit the time interval between the nearest neighbor and the initial phase point to be greater than the average orbital period of the time segment to avoid false proximity points. Record the initial distance between the initial phase point and the nearest neighbor. The phase point distance is traced along the phase point evolution trajectory after each time step Δt. Through formula Calculate the instantaneous Lyapunov exponent for each evolution step; The curves of all instantaneous Lyapunov exponents over time are fitted by linear regression using the least squares method. The slope of the fitted curve is the maximum Lyapunov exponent for that time segment. The maximum Lyapunov exponent of the two time series segments is compared with the preset threshold 0. Only when the maximum Lyapunov exponent of the two time series segments is greater than zero, the two time series segments are determined to have nonlinear chaotic time series characteristics that are strongly correlated with the topography and soil characteristics of the ridge and are combined into an effective fused time series set. If the judgment condition is not met, the length of the target sliding window is adjusted in a fixed step of ±20%. The upper and lower limits of the sliding window length are the number of sampling points corresponding to 5 insertion cycles and 0.5 insertion cycles, respectively. The maximum number of adjustments is no more than 5. After the adjustment is completed, the time series segment is re-trunculated and the maximum Lyapunov exponent calculation and characteristic verification are performed until a valid fused time series set that meets the requirements is obtained or the maximum number of adjustments is reached. Using the effectively fused time series set as input, the dimensionless statistics are calculated through correlation integrals. The delay time of the two time series segments is determined by the first minimum point of the dimensionless statistics curve, and the optimal time window is determined by the global minimum point of the dimensionless statistics curve. Then, the embedding dimension of the two time series segments is calculated according to the correlation formula between the optimal time window and the embedding dimension.

[0059] Specifically, using the effectively fused time series set as input, the CC algorithm is employed to calculate dimensionless statistics through correlation integrals. First, the effectively fused time series set is divided into t disjoint subsequences. For different embedding dimensions m, radial distance r, and time delays... The correlation integrals of the corresponding subsequences are calculated respectively. The correlation integral is the probability that the distance between any two phase points in the reconstructed phase space is less than the radial distance r. Based on the correlation integral, a dimensionless statistic for characterizing the correlation of time series is calculated. Plot dimensionless statistics over time delay The optimal delay time for the two sets of time segments is determined by the first minimum point of the changing curve. The optimal time window is determined by the global minimum point of the curve, and then based on the optimal time window... The correlation formula is used to calculate the optimal embedding dimension m of the two sets of time segments, where w represents the time window.

[0060] Using time delay and embedding dimension as input conditions, phase space reconstruction is performed on the effectively fused time series based on the temporal high-dimensional embedding criterion, mapping one-dimensional time series data to a high-dimensional phase space, and mining high-dimensional features that characterize the variation of ridge topography and soil planting resistance.

[0061] Among them, using delay time and embedding dimension as input conditions, phase space reconstruction is performed on the effectively fused time series set based on the temporal high-dimensional embedding criterion, mapping one-dimensional time series data to a high-dimensional phase space, and mining high-dimensional features that characterize the variation of ridge topography and soil planting resistance, including: Based on the delay time and embedding dimension of the two sets of time segments, high-dimensional embedding vectors of the two sets of time segments are constructed respectively. Based on the global unified time reference of the coordinated action timing instructions, the timestamp vectors of the two sets of high-dimensional embedded vectors are horizontally synchronously spliced ​​to generate a reconstructed phase space matrix of multivariate fusion. Based on the reconstruction of the phase space matrix by multivariate fusion, the evolution trajectory features of the state vector in the phase space are extracted, redundant noise components in the evolution trajectory features are removed, and only the high-dimensional feature set of terrain changes that can completely characterize the changes in ridge topography and soil planting resistance are retained. The high-dimensional feature set of terrain changes is sequentially subjected to false nearest neighbor verification and feature consistency verification. After the verification is passed, the high-dimensional features representing the terrain changes are obtained.

[0062] Specifically, the temporal high-dimensional embedding criterion is the delayed embedding rule of the Takens embedding theorem. The Takens embedding theorem states that for a nonlinear dynamic system with chaotic characteristics, as long as the reconstructed embedding dimension m satisfies m ≥ 2D + 1 (D is the fractal dimension of the original dynamic system), a high-dimensional phase space topologically equivalent to the original system can be reconstructed through delayed embedding of the time series, completely preserving all the dynamic characteristics of the original system. In this invention, the fused time-series data of driving acceleration and planting resistance are reconstructed using a multivariate split embedding and global stitching method. This fully exploits the temporal coupling characteristics between the two data streams, accurately characterizing the changing patterns of ridge topography and soil resistance. Specific steps include: Step 1: Using the preprocessed single-channel time series data (X1(t) / X2(t)), the corresponding delay time τ, and the embedding dimension m as input conditions, construct a high-dimensional embedding vector for each channel of time series data according to the delay embedding rule of the Tukens embedding theorem. For a time series sequence X(t) of length N, its effective embedding vector at time t is: Where t takes values ​​ranging from 1 to 2. To ensure that all delay components have valid measured data, with no null values ​​or extrapolated data, the above operations were performed on the driving acceleration time series X1(t) and the planting resistance time series X2(t) respectively, to obtain the high-dimensional embedding vector sets of the two time series, including the driving acceleration time series embedding vector set Y1(t) and the planting resistance time series embedding vector set Y2(t).

[0063] Step 2: Using the high-dimensional embedding vector sets Y1(t) and Y2(t) of the two time series as input conditions, and based on the global unified time reference of the cooperative action time series instructions, the simultaneous time stamp embedding vectors of the two time series are horizontally concatenated to construct a multivariate fusion reconstructed phase space matrix. In the concatenated reconstructed phase space, the fusion state vector at time t is: Y'(t)=[Y1(t),Y2(t)]=[X1(t),X1(t-τ1),…,X1(t-(m1-1)τ1),X2(t),X2(t-τ2),…,X2(t-(m2-1)τ2)]; This concatenation method completely preserves the dynamic evolution characteristics of the single time series, while also exploring the temporal coupling characteristics between driving acceleration and planting resistance, which is completely matched with the linkage change law of ridge topography undulation and soil planting resistance, and outputs the multivariate fusion reconstructed phase space matrix and the fusion state vector at each time.

[0064] Step 3: Using the multivariate fusion and reconstruction phase space matrix as input, based on the reconstructed high-dimensional phase space, extract the evolution trajectory features of the state vectors in the phase space, including Euclidean distance between phase points, trajectory evolution trend, and rate of change of adjacent phase points, which are highly correlated with the changes in ridge topography. At the same time, remove redundant noise components in the phase space, retain only the core feature components that can completely characterize the undulations of the ridge topography and changes in soil planting resistance, and output a high-dimensional feature set characterizing the topography changes. Step 4: Using the multivariate fusion reconstruction phase space matrix and the original time series data as input, verify the effectiveness of the reconstruction through two metrics to ensure that the reconstruction results meet the requirements of the Tukens embedding theorem: The False Nearest Neighbor (FNN) method is used for verification. When the proportion of false nearest neighbors in the reconstructed phase space is less than 5%, it is determined that the embedding dimension is reasonable and there is no topological distortion caused by insufficient dimension in the reconstructed phase space. The maximum Lyapunov exponent of the reconstructed phase space time series was recalculated, and the deviation from the maximum Lyapunov exponent of the original time series was no more than 10%, proving that the reconstructed phase space is topologically equivalent to the original nonlinear dynamic system and fully preserves the dynamic characteristics of the original system. The results of the reconstruction effectiveness verification and the high-dimensional characteristics of the terrain change were finally confirmed.

[0065] The high-dimensional features representing topographic changes are input into a pre-built weighted local prediction model. The model predicts the changing trends of ridge topographic undulation and soil planting resistance in future time periods and outputs ultra-short-term topographic prediction results.

[0066] Specifically, high-dimensional features representing topographic changes are input into a pre-constructed weighted local prediction model. This model then predicts the changing trends of ridge topographic undulation and soil planting resistance over future time periods and outputs ultra-short-term topographic prediction results, including: A weighted local prediction model is constructed based on the Euclidean distance between neighboring phase points in high-dimensional phase space and the local linear fitting rule. The original fused dataset, which includes the whole machine driving acceleration data and the planting resistance time series data, is divided into a training set and a validation set. The training set is input into the weighted local prediction model to perform iterative training. After each round of training, the weighted local prediction model is validated through the validation set. The trained weighted local prediction model is obtained when the maximum number of iterations is met.

[0067] Specifically, based on the high-dimensional phase space obtained by reconstructing the phase space, the core operation rules of the model are first determined: the model input is the current phase point in the high-dimensional phase space, and the output is the evolution trajectory of the phase point within a preset time period in the future; the number of neighboring phase points selected, k, is set to be 2 to 3 times the optimal embedding dimension m (for example, when m=6, k takes 12 to 18), the Euclidean distance between phase points is used as the quantitative index of dynamic similarity, and the reciprocal of the Euclidean distance is used as the weighting coefficient to perform first-order local linear fitting on the historical evolution trajectory of neighboring phase points, establish a linear mapping relationship from the current phase point coordinates to the future phase point coordinates, and complete the framework construction of the weighted local prediction model.

[0068] The original fusion dataset, which includes time-series data of the machine's acceleration and planting resistance, was collected through actual measurements and divided into a training set and a validation set in a 7:3 ratio. The training set data was reconstructed in phase space and then input into the model. The actual evolution trajectories of each historical phase point in the training set were used as labels to iteratively optimize the mapping coefficients of the local linear fitting. After each round of training, the model's prediction accuracy was verified using the validation set data, and the mean square error between the predicted and measured values ​​was calculated. When the mean square error was lower than the preset field operation threshold or the preset maximum number of iterations (30-50 times) was reached, training was stopped, and the weighted local prediction model was obtained.

[0069] The high-dimensional features characterizing topographic changes are input into the weighted local prediction model. The weighted local prediction model locates the high-dimensional phase point corresponding to the current high-dimensional feature in the reconstructed high-dimensional phase space, and selects the neighboring phase point whose dynamic characteristics are closest to that high-dimensional phase point.

[0070] Specifically, the high-dimensional features representing terrain changes extracted in real time are mapped to the reconstructed high-dimensional phase space to locate the high-dimensional phase point corresponding to the current moment (i.e., the multivariate fusion state vector at the current moment); the Euclidean distance between the current phase point and all historical phase points in the phase space is calculated, and the k nearest neighbor phase points are selected by sorting them from smallest to largest distance. At the same time, the time interval between the selected neighbor phase points and the current phase point is limited to not less than the optimal delay time τ to avoid interference from false neighbor points and ensure that the selected phase points are the effective phase points that are closest to the dynamic characteristics of the current working condition.

[0071] The weighted local prediction model uses the reciprocal of the Euclidean distance between the current high-dimensional phase point and its neighboring phase points as the weighting coefficients. Based on the preset local linear fitting rules, it completes the fitting calculation and obtains the evolution trajectory of the high-dimensional phase point within a preset time period in the future. The evolution trajectory of high-dimensional phase points is inversely mapped to the changing trend of ridge topography undulation and soil planting resistance, and the ultra-short-term topographic prediction results are output for duckbill mechanism and seedling placement mechanism.

[0072] Specifically, the weighting coefficients are based on the reciprocal of the Euclidean distance between the current phase point and each neighboring phase point. All weighting coefficients are normalized to ensure that the sum of all weights is 1, realizing the operation rule that "the closer the distance, the higher the weight and the greater the contribution to the prediction result". Based on the historical evolution data of k neighboring phase points, that is, the trajectory change pattern of each neighboring phase point from the corresponding time to the future prediction time, a weighted first-order local linear fitting is performed through the normalized weighting coefficients to obtain the fitted linear mapping matrix. This matrix is ​​the prediction model of the future evolution pattern of the current phase point.

[0073] The current phase point is calculated using the fitted linear mapping matrix to obtain the high-dimensional phase point evolution trajectory within a preset ultra-short time period (usually 0~200ms, corresponding to the distance the whole machine travels 3~5cm during field operations), which is the predicted phase point coordinates at each future moment. Then, through the inverse mapping of phase space reconstruction, the predicted trajectory in the high-dimensional phase space is restored to a one-dimensional whole machine travel acceleration time series and planting resistance time series, which are transformed into the future change trends of ridge topography undulation and soil planting resistance. Finally, the ultra-short-term terrain prediction results are output for the correction of the actions of the duckbill mechanism and the seedling feeding mechanism.

[0074] The ultra-short-term terrain prediction results are nonlinearly mapped to advance correction values ​​using a Kalman filter. Based on the advance correction values ​​and the timing instructions of the coordinated actions, the duckbill mechanism is driven to complete the adaptive contour insertion action. The duckbill mechanism is then driven to cooperate in completing the adaptive contour insertion action.

[0075] Specifically, the ultra-short-term terrain prediction results are nonlinearly mapped and random noise is filtered out by an unscented Kalman filter. The output of the duckbill mechanism's planting parameters and seedling timing advance correction amount, which match the field working conditions and are aligned with the global time reference of the whole machine, is then added to the coordinated action timing command in real time to complete dynamic correction. This drives the seedling placement mechanism and the duckbill mechanism to cooperate in a precise timing sequence to complete the adaptive contour planting action that adapts to the undulation of the ridge surface and changes in soil properties.

[0076] The transplanting execution parameter calibration and output module is used to sequentially perform filtering and noise reduction, outlier removal and dimensional calibration operations on the action execution and feedback data, generate standardized transplanting action parameters and output them to the ridge planting posture optimization unit.

[0077] The ridge planting posture optimization unit 5 is used to analyze action execution and feedback data, generate a full-domain adjustment command for ridge planting posture, update the machine's walking, steering and planting matching operating parameters according to the full-domain adjustment command for ridge planting posture, and send the updated operating parameters back to the machine control unit.

[0078] The full-range adjustment commands for ridge planting posture include: guiding driving commands along both sides of the ridge, adjustment commands to adapt to the height and parallelism of the ridge, adjustment commands to match the width of the ridge, and adjustment commands to adjust the spacing between seedlings.

[0079] Specifically, the ridge planting posture optimization unit 5 is the core of the closed-loop optimization of the whole machine transplanting operation. It receives standardized action execution and feedback data, analyzes and identifies the operation deviations of the whole machine driving centering, frame posture, ridge body adaptation, and seedling planting parameters, and generates corresponding full-domain adjustment instructions for ridge planting posture, including ridge guiding driving, ridge height and parallelism adjustment, ridge body width adaptation, and planting spacing correction. Then, it updates the core operating parameters of the whole machine walking, turning and planting matching according to the adjustment instructions, and sends the updated parameters back to the whole machine control unit 1, realizing adaptive closed-loop optimization of the whole process of tobacco seedling transplanting operation, and continuously improving the quality of field operation and the ability to adapt to complex working conditions.

[0080] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent planter for tobacco seedlings, characterized in that, The grower includes: The whole machine control unit is used to integrate the pre-acquired whole machine operating condition control signals and safety interlock control signals into whole machine control commands, and send the whole machine control commands to the motor set cooperative control unit; The motor set coordination control unit is used to allocate communication station numbers to the drive motor set based on the extended communication protocol to establish a motor communication addressing reference. The global time reference criterion established by the pulse coupling synchronization algorithm is used as a constraint condition. The drive motor set synchronization control command is generated according to the whole machine control command and the communication addressing reference. The walking drive and steering unit is used as the driving power source of the drive motor group. Based on the synchronous control command of the drive motor group, it coordinates the operation status between each drive motor to perform driving and steering actions, and synchronously outputs the driving status reference signal to the transplanting execution unit. The transplanting execution unit is used to generate coordinated action timing instructions that match the seedling feeding mechanism, the duckbill mechanism and the driving status reference signal according to the preset driving speed-planting frequency matching rules. The unit drives the seedling feeding mechanism to perform seedling feeding action and drives the duckbill mechanism to cooperate in completing adaptive contour planting action according to the coordinated action timing instructions. The unit acquires action execution and feedback data in real time during the seedling feeding and planting actions and transmits it to the ridge planting posture optimization unit. The ridge planting posture optimization unit is used to analyze action execution and feedback data, generate ridge planting posture global adjustment commands, update the machine's walking, steering and planting matching operating parameters according to the ridge planting posture global adjustment commands, and send the updated operating parameters back to the machine control unit.

2. The intelligent planter for tobacco seedlings according to claim 1, characterized in that, The ridge planting posture adjustment commands include: guide driving commands along both sides of the ridge, adjustment commands to adapt to the height and parallelism of the ridge, adjustment commands to match the width of the ridge, and adjustment commands to adjust the spacing between seedlings.

3. The intelligent planter for tobacco seedlings according to claim 1, characterized in that, The motor set collaborative control unit, when allocating communication station numbers to the drive motor set based on the extended communication protocol to establish a motor communication addressing reference, and using the global time reference criterion established by the pulse coupling synchronization algorithm as a constraint condition, generates synchronous control commands for the drive motor set based on the overall machine control commands and the communication addressing reference, includes: The extended communication protocol is configured using a dual verification mechanism that combines parity check and cyclic redundancy check. After the extended communication protocol is configured, a unique communication station number is assigned to each drive motor in the drive motor group. Establish a communication addressing benchmark that characterizes the mapping relationship between each drive motor and the communication station number; The target control parameters of each drive motor are obtained by parsing the overall machine control command. Based on the target control parameters of each drive motor, an initial motor speed command set is generated, which includes the angular velocity of the transplanter and the linear velocity of the walking motor and corresponds to the communication addressing reference. The pre-collected encoder feedback parameters, current loop sampling parameters, and adjacent drive motor state parameters are input into the Kalman filter for fusion calculation to obtain the actual operating state of each drive motor. Based on the actual operating state of each drive motor, the initial motor speed command set is corrected to obtain the corrected motor speed command set. A global time reference criterion is established using a pulse coupling synchronization algorithm. Under the constraints of the global time reference criterion, the coupling strength between each drive motor is adjusted according to the corrected motor speed command set, and a phase-aligned motor group synchronization control command is output.

4. The intelligent planter for tobacco seedlings according to claim 3, characterized in that, The process of establishing a global time reference criterion using a pulse coupling synchronization algorithm, and adjusting the coupling strength between each drive motor based on the corrected motor speed command set under the constraints of the global time reference criterion, outputting phase-aligned motor group synchronization control commands includes: For each drive motor in the drive motor group, a corresponding pulse coupling oscillator is constructed, and a unified counting period and reference beat are set for each pulse coupling oscillator. A global time reference criterion for synchronization of each drive motor is established using the reference beat generator as a carrier. Based on the global time base criterion and the initial state parameters of the pulse-coupled oscillators, a preset linear phase function is used to periodically accumulate and update the phase state of each pulse-coupled oscillator to obtain the updated phase state of each pulse-coupled oscillator. The update phase state of each pulse-coupled oscillator is monitored in real time. When the update phase state of any pulse-coupled oscillator meets the preset phase excitation threshold, the pulse-coupled oscillator is determined to enter the excitation state. After the pulse-coupled oscillator completes the excitation action, the corresponding excitation state signal and coupling strength parameter are output. The excitation state signal and the corresponding coupling strength parameter are broadcast to the non-excited pulse-coupled oscillator, and the phase state of the non-excited pulse-coupled oscillator is updated by phase increment using the phase formula. The coupling strength between each pulse-coupled oscillator is dynamically adjusted based on the phase state of the pulse-coupled oscillator updated by the phase increment and in conjunction with the corrected motor speed command set. Based on the coupling strength between each pulse-coupled oscillator, a shielding period verification is performed on each pulse-coupled oscillator, and a phase-aligned synchronous control command for the motor set is generated based on the shielding period verification result.

5. The intelligent planter for tobacco seedlings according to claim 4, characterized in that, The process of performing a shielding period check on each pulse-coupled oscillator based on the coupling strength between them, and generating phase-aligned motor synchronization control commands based on the shielding period check results, includes: After the pulse-coupled oscillator completes the excitation action, a shielding period of preset duration is initiated based on the coupling strength between each pulse-coupled oscillator. During the shielding period, each pulse-coupled oscillator maintains its own periodic cumulative update mechanism, shields itself from receiving the coupling strength parameters of other pulse-coupled oscillators, and temporarily does not perform phase increment updates; after the shielding period ends, phase increment updates are resumed, and the phase naturally accumulated during the shielding period is used as the effective phase parameter output. Based on the effective phase parameters of the output, compare the real-time phase deviation of all pulse-coupled oscillators; When the phase deviation is within the synchronization convergence threshold range, it is determined that the drive motor group has achieved phase alignment of multiple drive motors, and the motor group synchronization control command is generated based on the aligned unified phase and the corrected motor speed command set; otherwise, the phase accumulation update and synchronization convergence process is re-executed.

6. The intelligent planter for tobacco seedlings according to claim 1, characterized in that, The transplanting execution unit includes a driving status perception module, a collaborative instruction generation module, and a seedling placement collaborative drive module. Among them, the driving status perception module is used to analyze and extract the real-time driving speed and driving phase reference of the whole vehicle based on the driving status synchronization reference signal; The collaborative instruction generation module is used to convert the real-time driving speed and the walking phase reference into collaborative action timing instructions for the seedling dispensing mechanism and the duckbill mechanism that are linked to the driving state, based on the driving speed-planting frequency matching rule. The seedling feeding collaborative drive module is used to drive the seedling feeding mechanism and the duckbill mechanism to move sequentially according to the collaborative action timing instructions through the terrain adaptive contouring algorithm. During this process, action execution and feedback data are collected in real time. The transplanting execution parameter calibration and output module is used to sequentially perform filtering and noise reduction, outlier removal and dimensional calibration operations on the action execution and feedback data, generate standardized transplanting action parameters and output them to the ridge planting posture optimization unit.

7. The intelligent planter for tobacco seedlings according to claim 1, characterized in that, The seedling dispensing collaborative drive module, when driving the seedling dispensing mechanism and the duckbill mechanism sequentially according to the collaborative action timing instructions using a terrain adaptive contouring algorithm, includes: Real-time data collection of machine acceleration and planting resistance time-series data is used to construct a fused dataset. Nonlinear temporal characteristics representing ridge topographic undulation and soil planting resistance are extracted from the fused dataset, and phase space reconstruction is performed on the fused dataset based on the nonlinear temporal characteristics to obtain high-dimensional features representing topographic changes. The high-dimensional features representing topographic changes are input into a pre-built weighted local prediction model. The weighted local prediction model is used to predict the changing trends of ridge topographic undulation and soil planting resistance in future time periods and output ultra-short-term topographic prediction results. The ultra-short-term terrain prediction results are nonlinearly mapped to advance correction values ​​using a Kalman filter. Based on the advance correction values ​​and the timing instructions of the coordinated actions, the duckbill mechanism is driven to complete the adaptive contour insertion action. The duckbill mechanism is then driven to cooperate in completing the adaptive contour insertion action.

8. The intelligent planter for tobacco seedlings according to claim 7, characterized in that, The process involves extracting nonlinear temporal characteristics representing ridge topographic undulation and soil planting resistance from the fused dataset, and performing phase space reconstruction on the fused dataset based on these nonlinear temporal characteristics to obtain high-dimensional features representing topographic changes, including: Extract the corresponding length of the whole machine driving acceleration time series and planting resistance time series from the fused dataset according to the target sliding window; Calculate the maximum Lyapunov exponent of the two time series segments. Only when the maximum Lyapunov exponent of both time series segments is greater than zero is the two time series segments determined to have nonlinear time series characteristics and combined into an effective fused time series set. If the determination condition is not met, return to adjust the length of the target sliding window. Using the effectively fused time series set as input, the dimensionless statistics are calculated through correlation integrals; the delay time of the two time series segments is determined by the first minimum point of the dimensionless statistics curve, the optimal time window is determined by the global minimum point of the dimensionless statistics curve, and the embedding dimension of the two time series segments is calculated according to the correlation formula between the optimal time window and the embedding dimension. Using time delay and embedding dimension as input conditions, phase space reconstruction is performed on the effectively fused time series based on the temporal high-dimensional embedding criterion, mapping one-dimensional time series data to a high-dimensional phase space, and mining high-dimensional features that characterize the variation of ridge topography and soil planting resistance.

9. An intelligent planter for tobacco seedlings according to claim 8, characterized in that, The process of using delay time and embedding dimension as input conditions, and performing phase space reconstruction on the effectively fused time series set based on the temporal high-dimensional embedding criterion, maps one-dimensional time series data to a high-dimensional phase space, and extracts high-dimensional features that characterize the variation of ridge topography and soil planting resistance from it, including: Based on the delay time and embedding dimension of the two sets of time segments, high-dimensional embedding vectors of the two sets of time segments are constructed respectively. Based on the global unified time reference of the coordinated action timing instructions, the timestamp vectors of the two sets of high-dimensional embedded vectors are horizontally synchronously spliced ​​to generate a reconstructed phase space matrix of multivariate fusion. Based on the reconstruction of the phase space matrix by multivariate fusion, the evolution trajectory features of the state vector in the phase space are extracted, redundant noise components in the evolution trajectory features are removed, and only the high-dimensional feature set of terrain changes that can completely characterize the changes in ridge topography and soil planting resistance is retained. The high-dimensional feature set of terrain changes is sequentially subjected to false nearest neighbor verification and feature consistency verification. After the verification is passed, the high-dimensional features representing the terrain changes are obtained.

10. An intelligent planter for tobacco seedlings according to claim 9, characterized in that, The process of inputting high-dimensional features characterizing terrain changes into a pre-constructed weighted local prediction model, and predicting the changing trends of ridge topography undulation and soil planting resistance in future time periods through the weighted local prediction model, and outputting ultra-short-term terrain prediction results includes: A weighted local prediction model is constructed based on the Euclidean distance between neighboring phase points in high-dimensional phase space and the local linear fitting rule. The original fusion dataset containing the whole machine driving acceleration data and planting resistance time series data is divided into a training set and a validation set. The training set is input into the weighted local prediction model to perform iterative training. After each round of training, the weighted local prediction model is validated through the validation set. The trained weighted local prediction model is obtained when the maximum number of iterations is met. The high-dimensional features characterizing topographic changes are input into the weighted local prediction model. The weighted local prediction model locates the high-dimensional phase point corresponding to the current high-dimensional feature in the reconstructed high-dimensional phase space and selects the neighboring phase points with the dynamic characteristics of the high-dimensional phase point. The weighted local prediction model uses the reciprocal of the Euclidean distance between the current high-dimensional phase point and its neighboring phase points as the weighting coefficients. Based on the preset local linear fitting rules, it completes the fitting calculation and obtains the evolution trajectory of the high-dimensional phase point within a preset time period in the future. The evolution trajectory of high-dimensional phase points is inversely mapped to the changing trend of ridge topography undulation and soil planting resistance, and the ultra-short-term topographic prediction results are output for duckbill mechanism and seedling placement mechanism.

Images