Remote new energy tractor intelligent control method and system

Abstract

The invention relates to a remote new energy tractor intelligent control method and system. According to the method, the microstructure, stress parameters and humidity of soil are obtained, and dynamic soil cohesion and friction angle are generated through Kalman filtering fusion; a friction coefficient mapping table is inquired based on the parameters to construct a steering resistance torque model, the friction coefficient is calibrated through the cloud in combination with the actual measurement value of the torque sensor, and a resistance torque calibration value is generated; inputting the calibrated resistance moment and path tracking error into a fuzzy reasoning model, and dynamically adjusting parameters of a PI (Proportional-Integral-Definition) controller; a target steering angle is generated based on the corrected PI D parameters, a steering motor is driven to execute through a control algorithm, and state data of the tractor are collected; and uploading the state data to the cloud, and optimizing the friction coefficient mapping table through reinforcement learning. Prediction of steering resistance in a complex soil environment and adaptive matching of control parameters are realized, path tracking precision and system stability are remarkably improved, energy consumption is reduced, and safety and reliability of remote control are ensured.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent control technology, and in particular relates to a remote new energy tractor intelligent control method and system. Background Art

[0002] Current steering control systems for new energy tractors mostly use fixed-parameter control strategies, such as relying on preset PID parameters for path tracking. However, in complex soil environments, dynamic characteristics such as soil hardness and moisture can cause drastic changes in steering resistance, leading to path tracking deviations. Traditional methods fail to dynamically link soil mechanical properties with controller gains, especially in remote control scenarios where adaptive adjustment is inefficient when edge computing capabilities are limited. Furthermore, existing agricultural machinery path tracking technologies generally use a single control model, lacking a real-time fusion and verification mechanism for multi-source sensor data. This model's generalization capabilities are insufficient, making it difficult to adapt to steering requirements in diverse terrains (such as hills and muddy farmland).

[0003] While existing research has attempted to improve path tracking accuracy through fuzzy control or model prediction methods, these solutions often overlook the dynamic modeling of soil-vehicle interaction forces or fail to address communication latency and command security issues during multi-machine collaborative operations. For example, parameter tuning for some control systems relies on manual trial and error, which is time-consuming and difficult to adapt to dynamic environments. Other solutions, while incorporating cloud-based optimization, fail to achieve closed-loop edge-cloud collaboration, resulting in insufficient real-time control instructions. Furthermore, traditional steering resistance models fail to consider online identification of soil parameters (such as cohesion and friction angle), limiting the accuracy of steering torque prediction and making it difficult to accurately support dynamic adjustment of PID parameters.

[0004] To address the above issues, there is an urgent need for an intelligent control method that can integrate soil dynamic characteristics in real time, adaptively adjust control parameters, and balance safety and efficiency in remote scenarios. Summary of the Invention

[0005] Based on this, it is necessary to provide a remote new energy tractor intelligent control method and system to address the above technical problems.

[0006] In a first aspect, the present application provides a remote new energy tractor intelligent control method, comprising:

[0007] S1. Acquire soil micromorphology data, stress parameter data, and moisture data, and fuse the micromorphology data, stress parameter data, and moisture data through Kalman filtering to obtain soil cohesion and friction angle;

[0008] S2. Based on the soil cohesion and friction angle, the initial friction coefficient is retrieved from a preset friction coefficient mapping table on the cloud. Based on the soil cohesion, friction angle, and initial friction coefficient, a steering resistance torque model is constructed on the locally deployed edge computing node.

[0009] S3. Based on the measured operating parameters of the tractor, the initial friction coefficient is calibrated on the cloud to obtain a calibrated friction coefficient. The steering resistance torque model of the edge computing node is updated based on the calibrated friction coefficient to obtain a calibrated steering resistance torque.

[0010] S4, according to the calibrated steering resistance torque and path tracking error, using the fuzzy inference model to correct the PID parameters in real time to obtain corrected PID parameters;

[0011] S5. Generate a target steering angle based on the corrected PID parameters; and drive the steering motor to perform steering according to the target steering angle through a control algorithm, and collect state data of the tractor during the steering process;

[0012] S6. Feedback the status data to the cloud, and use reinforcement learning to optimize the friction coefficient mapping table based on the status data.

[0013] In a second aspect, the present application also provides a remote new energy tractor intelligent control system, which includes: a data acquisition module, a control module, a cloud server, and a locally deployed edge computing node; the cloud server includes a friction coefficient query module and a friction coefficient calibration module; the edge computing node includes a Kalman filter fusion module, a model construction module, a model update module, a PID parameter correction module, and a control parameter generation module;

[0014] A data acquisition module is used to obtain soil micromorphology data, stress parameter data, and moisture data;

[0015] A Kalman filter fusion module is used to receive the micro-morphology data, stress parameter data and humidity data sent by the data acquisition module, and fuse the micro-morphology data, stress parameter data and humidity data through Kalman filtering to obtain soil cohesion and friction angle;

[0016] The friction coefficient query module is used to receive the soil cohesion and friction angle sent by the Kalman filter fusion module, and query the initial friction coefficient through a preset friction coefficient mapping table based on the soil cohesion and friction angle;

[0017] The model building module is used to receive the initial friction coefficient sent by the friction coefficient query module and the soil cohesion and friction angle sent by the Kalman filter fusion module; and build a steering resistance torque model based on the soil cohesion, friction angle and initial friction coefficient;

[0018] a friction coefficient calibration module, configured to receive the measured operating parameters of the tractor uploaded by the data acquisition module, and calibrate the initial friction coefficient based on the measured operating parameters to obtain a calibrated friction coefficient;

[0019] a model updating module, configured to receive the calibrated friction coefficient sent by the friction coefficient calibration module, and update the steering resistance torque model according to the calibrated friction coefficient to obtain the calibrated steering resistance torque;

[0020] a PID parameter correction module, configured to receive the calibrated steering resistance torque sent by the steering resistance torque model update module, and to correct the PID parameters in real time using a fuzzy inference model according to the calibrated steering resistance torque and the path tracking error, thereby obtaining corrected PID parameters;

[0021] a control parameter generation module, configured to receive the corrected PID parameters sent by the PID parameter correction module, and generate a target steering angle based on the corrected PID parameters; and generate a control instruction through a control algorithm according to the target steering angle;

[0022] The control module is used to receive the control instructions sent by the control parameter generation module, drive the steering motor to perform steering according to the control instructions, and trigger the data acquisition module to collect the status data of the tractor during the steering process;

[0023] The friction coefficient query module is also used to receive the status data uploaded by the data acquisition module, and use reinforcement learning to optimize the friction coefficient mapping table according to the status data.

[0024] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, a remote new energy tractor intelligent control method as in the first aspect is implemented.

[0025] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a remote new energy tractor intelligent control method as in the first aspect.

[0026] The aforementioned remote intelligent control method and system for new energy tractors obtains soil micromorphology data, stress parameter data, and moisture data, and generates dynamic soil cohesion and friction angle through Kalman filtering. A steering resistance torque model is constructed based on these parameters by querying a friction coefficient mapping table. The friction coefficient is calibrated in the cloud using the torque sensor's measured values to generate a resistance torque calibration value. The calibrated resistance torque and path tracking error are input into a fuzzy inference model to dynamically adjust the proportional, integral, and differential parameters of the PID controller. A target steering angle is generated based on the corrected PID parameters, and the steering motor is driven by a control algorithm. The tractor's status data during the steering process is collected. Finally, this status data is uploaded to the cloud, and the friction coefficient mapping table is optimized through reinforcement learning. This enables real-time and accurate prediction of steering resistance and adaptive matching of control parameters in complex soil environments, significantly improving path tracking accuracy and system stability, reducing energy consumption, and ensuring the safety and reliability of remote control. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0028] Figure 1 A schematic flow chart of a remote intelligent control method for a new energy tractor provided by the present invention;

[0029] Figure 2 Schematic diagram of a flow chart for calibrating steering resistance torque in an optional embodiment of the present invention;

[0030] Figure 3 This is a structural schematic diagram of a remote new energy tractor intelligent control system provided by the present invention. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0032] refer to Figure 1 , which presents a flow chart of a remote new energy tractor intelligent control method provided by the present application, the method comprising the following steps:

[0033] S1. Obtain soil micromorphology data, stress parameter data, and moisture data, and fuse the micromorphology data, stress parameter data, and moisture data through Kalman filtering to obtain soil cohesion and friction angle.

[0034] Specifically, to obtain soil micromorphological data, high-resolution laser scanning sensors or optical sensors can be used. These sensors can be mounted on the front of the tractor, close to the ground, enabling them to accurately capture microscopic features such as soil particle size, shape, and surface texture. Stress parameter data can be collected using pressure sensors installed at the contact point between the tractor's tires or tracks and the soil. These sensors can sense in real time the stress distribution generated in the soil by the tractor's rolling motion. Moisture data can be obtained using soil moisture sensors, whose probes are inserted into the soil at a certain depth to measure the soil's moisture content.

[0035] When using Kalman filtering for multi-source data fusion, multiple Kalman filters run in parallel, each processing a different soil parameter to initially remove noise. These pre-processed data are then fed into a central Kalman filter, which comprehensively considers the correlations between the various parameters and performs deep fusion, ultimately yielding accurate estimates of soil cohesion and friction angle.

[0036] S2. Based on the soil cohesion and friction angle, the initial friction coefficient is queried on the cloud using a preset friction coefficient mapping table. Based on the soil cohesion, friction angle, and initial friction coefficient, a steering resistance torque model is constructed on the locally deployed edge computing node.

[0037] Specifically, the friction coefficient mapping table in the cloud is constructed based on a large amount of soil experimental data. By conducting tractor driving tests under different soil types and humidity conditions, the soil parameters and the corresponding friction coefficient between the tire and the soil are recorded, and a mapping relationship is established. The edge computing node can be composed of a high-performance embedded computing unit installed on the tractor, which has real-time data processing and model building capabilities. The construction of the steering resistance torque model is based on classical soil mechanics theory, such as Coulomb's friction law and Mohr-Coulomb criterion. It comprehensively considers soil cohesion, friction angle and initial friction coefficient, and obtains the functional relationship between the steering resistance torque and these parameters through mathematical modeling. The model form can be a polynomial function, exponential function, etc. The specific form can be determined by fitting experimental data, and then the model parameters are determined using parameter estimation methods such as the least squares method.

[0038] S3. Based on the measured operating parameters of the tractor, the initial friction coefficient is calibrated on the cloud to obtain the calibrated friction coefficient. The steering resistance torque model of the edge computing node is updated according to the calibrated friction coefficient to obtain the calibrated steering resistance torque.

[0039] Specifically, the measured operating parameters of the tractor include vehicle speed, acceleration, steering angle, steering wheel angular rate, engine speed, torque, etc. These parameters are collected in real time by various sensors on the tractor and transmitted to the cloud through a wireless communication module. The calibration process in the cloud can use an optimization algorithm, such as a particle swarm optimization algorithm or a genetic algorithm. For example, the measured operating parameters and the steering resistance torque measurement under current soil conditions are used as input, and the initial friction coefficient is iteratively optimized with the goal of minimizing the error between the predicted steering resistance torque and the actual measurement value. The optimized friction coefficient is fed back to the edge computing node through wireless communication, and the edge computing node uses this coefficient to update the steering resistance torque model. The update process involves recalculating the model parameters or locally adjusting the model structure to ensure the accuracy and reliability of the model under the current working conditions.

[0040] S4. According to the calibrated steering resistance torque and path tracking error, the PID parameters are corrected in real time using the fuzzy inference model to obtain corrected PID parameters.

[0041] Specifically, the construction of the fuzzy inference model is based on expert experience and extensive experimental data. First, the fuzzy sets of the input variables (calibrated steering torque and path tracking error) and the output variables (PID parameter corrections) are determined, including fuzzy subsets such as "small," "medium," and "large." Then, a fuzzy rule base is developed, such as "If the steering torque is large and the path tracking error is large, increase the PID proportional gain," with each rule corresponding to a fuzzy relationship. When the input data enters the fuzzy inference model, it is first fuzzified to convert the precise input value into a fuzzy membership value. Then, reasoning is performed based on the fuzzy rules. Mamdani-type fuzzy inference methods can be used to obtain the output fuzzy set through fuzzy synthesis. Finally, defuzzification (such as the center of gravity method) is used to obtain the precise PID parameter corrections, enabling real-time dynamic adjustment of the PID parameters.

[0042] S5. Generate a target steering angle based on the corrected PID parameters; and according to the target steering angle, drive the steering motor to perform steering through a control algorithm, and collect state data of the tractor during the steering process.

[0043] Specifically, a classic PID control algorithm is used to generate the target steering angle based on the corrected PID parameters. The control algorithm calculates the required target steering angle based on the current path tracking error and the PID parameters, ensuring that the tractor's trajectory remains as close to the preset path as possible. The target steering angle is achieved through the steering motor drive system, which consists of an electric motor, a reducer, and a steering mechanism. The electric motor generates torque based on the control signal, which is amplified by the reducer and drives the steering mechanism to achieve the tractor's steering motion. During the steering process, an encoder mounted on the steering motor collects real-time data on motor angle and speed. Simultaneously, the vehicle attitude sensor acquires information on the tractor's yaw rate, lateral acceleration, and other status information. This data is used for subsequent feedback control and model optimization.

[0044] S6. Feedback the status data to the cloud, and use reinforcement learning to optimize the friction coefficient mapping table based on the status data.

[0045] Specifically, after state data is fed back to the cloud, a reinforcement learning algorithm evaluates the performance of the current friction coefficient mapping table based on a preset reward function, such as path tracking accuracy and steering stability. A reinforcement learning model (such as a Q-learning algorithm or a deep reinforcement learning network) updates the parameters in the mapping table based on the current state and reward signal, continuously optimizing the mapping relationship. This optimized friction coefficient mapping table improves the accuracy of initial friction coefficient queries, thereby enhancing the performance of the entire control system.

[0046] The aforementioned remote intelligent control method for new energy tractors obtains soil micromorphology data, stress parameter data, and moisture data, and generates dynamic soil cohesion and friction angle through Kalman filtering. A steering resistance torque model is constructed by querying a friction coefficient mapping table based on these parameters. The friction coefficient is calibrated in the cloud using the torque sensor's measured values to generate a resistance torque calibration value. The calibrated resistance torque and path tracking error are input into a fuzzy inference model to dynamically adjust the proportional, integral, and differential parameters of the PID controller. A target steering angle is generated based on the corrected PID parameters, and the steering motor is driven by a control algorithm. The tractor's state data during the steering process is collected. Finally, this state data is uploaded to the cloud, and the friction coefficient mapping table is optimized through reinforcement learning. This method enables real-time and accurate prediction of steering resistance and adaptive matching of control parameters in complex soil environments, significantly improving path tracking accuracy and system stability, reducing energy consumption, and ensuring the safety and reliability of remote control.

[0047] In an optional embodiment, S1 includes the following steps:

[0048] S11. Scan the surface morphology through LiDAR and extract the local terrain slope of the soil as microscopic morphological data.

[0049] Specifically, the surface topography is scanned by a laser radar installed at the front of the tractor. The laser radar emits laser pulses at a certain frequency and receives reflected light, completing dense sampling of the local terrain in front of the tractor in a very short time. The three-dimensional coordinates of each sampling point constitute high-precision point cloud data. Based on these point cloud data, the local terrain slope of the soil is extracted using a terrain analysis algorithm. The core of this algorithm is to construct a local terrain fitting surface. With each laser sampling point as the center, several surrounding neighboring points are selected and the surface equation is fitted using the least squares method. The slope value at that point can be calculated based on the partial derivatives of the surface equation. The local terrain slope matrix of the entire scanning area is calculated in sequence according to the set spatial resolution. The slope value quantifies the degree of inclination of the soil surface, characterizes the terrain undulation characteristics, and provides key basic data for subsequent soil mechanical property analysis.

[0050] S12. Using a soil triaxial sensor, the compressive stress and shear stress are measured as stress parameter data.

[0051] Specifically, the triaxial soil sensor is buried in the soil depth range where tractors frequently operate. Its measurement core is based on strain gauge pressure sensing technology. The elastic element inside the sensor generates micro-strain under the action of soil compressive stress and shear stress. The strain gauge attached to the surface of the elastic element then changes its resistance value. The four bridge arms of the bridge circuit are excited by a DC power supply and output a millivolt voltage signal proportional to the stress. After the signal is amplified by an amplifier and filtered to remove high-frequency noise, it is converted into a digital signal by an analog-to-digital converter and transmitted to the data acquisition terminal. The collected compressive stress and shear stress data series accurately reflect the mechanical response state of the soil under the tractor's operating load and are the key raw basis for the subsequent evaluation of soil strength characteristics and the construction of a steering resistance model.

[0052] S13, receiving infrared band data from the satellite, and calculating soil moisture data through normalized differential moisture index.

[0053] Specifically, the satellite data receiving terminal equipped at the ground station receives in real time multispectral remote sensing data including infrared bands sent by meteorological satellites passing over the area. For soil moisture inversion, the focus is on extracting the reflectance data of the infrared band and calculating it using the normalized difference moisture index (NDWI) model. NDWI is defined as the difference between the reflectance of the near-infrared band and the reflectance of the short-wave infrared band divided by the sum of the two, and its value is significantly positively correlated with the soil moisture content. Based on the calibration equation established based on the measured humidity of the soil samples in the early stage and the NDWI value, the calculated NDWI is converted into a specific soil moisture value to realize dynamic monitoring of the soil moisture in the tractor operation area. The humidity data is used as a key environmental variable in the intelligent control system to correct soil mechanical parameters and optimize the control strategy.

[0054] S14. Perform Kalman filtering iteration on the local terrain slope, bearing stress, shear stress, and humidity data to generate dynamic soil cohesion and friction angle. The expressions for soil cohesion and friction angle are:

[0055]

[0056] Where C is the soil cohesion, φ is the friction angle, τ s is the shear stress, is the maximum value of shear stress, σ p is the compressive stress, ω soil is the humidity data, γ is the local terrain slope, k0 is the humidity correction factor, Δφ(γ,ω soil ) is the terrain-humidity compensation function.

[0057] Specifically, the local terrain slope, bearing stress, shear stress, and humidity data obtained in steps S11 to S13 are constructed as a state vector and input into the Kalman filter. The initial state estimation of the filter is based on the historical average soil parameter setting, and the prediction equation is derived according to the basic model of soil mechanics. The state transfer matrix is constructed by considering the prior knowledge that the soil cohesion and friction angle change with stress, humidity, and slope. The update equation is based on the current measurement data, calculates the Kalman gain, corrects the predicted state, and iteratively updates the estimated values of soil cohesion and friction angle. The calculation formula for soil cohesion is: Where C is the soil cohesion, which reflects the bonding strength between soil particles; τ s is the shear stress, reflecting the ability of soil to resist shear failure; σ p is the compressive stress, caused by the pressure of the tractor on the soil; ω soil is the humidity data, which affects the water film thickness and cohesion between soil particles; k0 is the humidity correction factor, which can be determined through a large number of experimental fittings to quantify the weakening effect of humidity on cohesion; φ is the friction angle, which represents the friction resistance between soil particles. The calculation formula of the friction angle is: in, is the maximum shear stress, corresponding to the ultimate shear strength of the soil under the current stress conditions; γ is the local terrain slope, which can change the soil stress distribution and affect the contact state between particles; Δφ(γ,ω soil ) is the terrain-humidity compensation function, which can be obtained by fitting the experimental data to comprehensively correct the deviation of the terrain slope and humidity on the friction angle.

[0058] In an optional embodiment, the steering resistance torque model is expressed as:

[0059]

[0060] Among them, T ris the steering resistance torque, μ0 is the initial friction coefficient, R w is the wheel radius of the tractor, k s is the slip attenuation factor; S is the wheel slip rate, defined as v is the tractor's travel speed, and ω is the tractor's wheel angular velocity.

[0061] Specifically, the slip attenuation factor k s It is a parameter obtained by fitting experimental data and is used to describe the nonlinear attenuation effect of wheel slip on the friction coefficient. s To find the value of , we can conduct tractor driving tests under different soil conditions and slip rates, record the wheel slip rate S and the corresponding friction coefficient μ, and then use the nonlinear regression method to fit the data to the expression So we can get the optimal k s value and stores it in the edge computing node.

[0062] The slip rate S is calculated by the tractor's forward speed v and the wheel angular velocity ω, and its formula is: When the tractor is operating, a speed sensor mounted on the tractor body measures the forward speed v in real time, while a wheel encoder measures the wheel angular velocity ω. The edge computing node calculates the slip rate S based on this real-time data and uses it as an input parameter for calculating the steering resistance torque.

[0063] refer to Figure 2 In an optional embodiment, S3 includes the following steps:

[0064] S31. Collect the measured value of the tractor's torque sensor, calculate the initial steering resistance torque using the steering resistance torque model, and upload the initial steering resistance torque and the measured value of the torque sensor to the edge computing node.

[0065] Specifically, torque sensors are installed at two key locations in the steering system: the steering shaft and the steering motor output shaft. These torque sensors are based on the strain gauge measurement principle. Their core component is a strain gauge bonded to an elastic element. When torque acts on the elastic element, it generates a tiny strain, causing the strain gauge's resistance to change. This resistance change is converted into a voltage signal using a Wheatstone bridge circuit. After amplification by an amplifier and noise filtering by a filter, it is converted into a digital signal by an analog-to-digital converter. During installation, the torque sensor is connected to the steering shaft and the motor output shaft via high-precision couplings to ensure accurate torque transmission. The signal cables are carefully routed and shielded to minimize electromagnetic interference and ensure signal stability and reliability. After installation, the sensor is calibrated, which can be performed in the factory. This process establishes a precise correlation between torque and signal by applying a known torque value and recording the sensor's output signal.

[0066] At the same time, the edge computing node uses the previously constructed steering torque model to calculate an initial theoretical steering torque value based on current soil conditions and tractor operating parameters. This initial theoretical steering torque value, along with the actual value collected by the torque sensor, is uploaded to the edge computing node via the wireless communication module.

[0067] S32. On the edge computing node, compare the initial steering resistance torque with the actual value measured by the torque sensor. If the relative error is greater than the preset error threshold, the cloud is triggered to calibrate the initial friction coefficient, obtain the calibrated friction coefficient, and send the calibrated friction coefficient to the edge computing node; where the relative error is T real is the actual value measured by the torque sensor.

[0068] Specifically, the initial steering resistance torque theoretical value and the actual value of the torque sensor are compared and analyzed on the edge computing node, that is, the relative error between the two is calculated. The formula is: Where T r is the initial steering resistance torque calculated by the model, T real is the actual value collected by the torque sensor. This relative error reflects the degree of deviation between the model's predicted value and the actual value. If this relative error exceeds the preset error threshold, it indicates that the current initial friction coefficient μ0 no longer accurately describes the friction characteristics between the soil and the wheel, and calibration is required. The edge computing node then triggers the calibration process in the cloud.

[0069] The cloud, leveraging its computing power and the vast amount of historical data it stores, uses an optimization algorithm to recalculate and calibrate the initial friction coefficient, deriving a calibrated friction coefficient μ′ that better reflects the current operating conditions. This new friction coefficient is then transmitted back to the edge computing node via a wireless communication network.

[0070] S32. On the edge computing node, the steering resistance torque model is updated according to the calibrated friction coefficient to obtain the calibrated steering resistance torque; the expression of the updated steering resistance torque model is:

[0071]

[0072] Among them, T′ r is the steering resistance torque after calibration, and μ′ is the friction coefficient after calibration.

[0073] Specifically, after the edge computing node receives the calibrated friction coefficient μ′, it updates the steering resistance torque model according to the new friction coefficient. The updated model expression becomes Among them, T′ rrepresents the calibrated steering resistance torque. This update process is more than just a simple parameter replacement; it involves adaptively adjusting the model to current soil conditions and tractor operating status. This allows the steering resistance torque model to dynamically reflect the tractor's actual steering resistance under varying soil conditions in real time, providing more accurate data support for subsequent control strategies and ensuring more precise and efficient tractor steering control.

[0074] In an optional embodiment, triggering the cloud to calibrate the initial friction coefficient to obtain the calibrated friction coefficient, and sending the calibrated friction coefficient to the edge computing node includes the following steps:

[0075] S321. Establish a dynamic digital twin model of the interaction between tractor and soil based on the physics engine.

[0076] Specifically, a cloud-based physics engine builds a dynamic digital twin model of the interaction between the tractor and the soil. This model accurately simulates the tractor's steering behavior under different soil conditions. Its core principle is to digitally model the tractor's mechanical structure and soil mechanical properties to achieve a high degree of fidelity to the real operating environment. The physics engine uses Newtonian mechanics and soil mechanics equations to calculate the forces and motion of the tractor under various operating conditions.

[0077] S322. Inject the actual measured value of the torque sensor into the dynamic digital twin model, and iteratively adjust the initial friction coefficient until the absolute value of the difference between the simulated value of the steering resistance torque in the simulation result output by the dynamic digital twin model and the actual measured value of the torque sensor is less than or equal to the preset value, thereby obtaining the calibrated friction coefficient.

[0078] Specifically, the cloud platform runs an optimization algorithm through a combination of simulation and actual measurements, iterating multiple times on the initial friction coefficient. Each iteration simulates the tractor's steering process under the same operating conditions, outputting a simulated steering resistance torque value, which is then compared with the actual torque sensor measured value. When the absolute value of the difference between the simulation result and the actual value is less than or equal to a preset accuracy threshold, the current friction coefficient is considered to have converged to the optimal value, thus obtaining the calibrated friction coefficient.

[0079] S323. The calibrated friction coefficient is encrypted and transmitted to the edge computing node via the wireless network.

[0080] Specifically, to ensure data security, the calibrated friction coefficient is transmitted back to the edge computing node via a wireless network using an encrypted protocol. Upon receiving the data, the edge computing node decrypts and verifies it to ensure its integrity and accuracy. This data is then applied to the steering resistance torque model, improving its adaptability to current soil conditions and its prediction accuracy.

[0081] In an optional embodiment, S4 includes the following steps:

[0082] S41, calculate the torque deviation according to the calibrated steering resistance torque, and calculate the path tracking angle error at the same time; wherein the torque deviation is ΔT r =T′ r -T real , the path tracking angle error is e θ =θ cmd -θ act ,θ cmd is the target steering angle of the planned path, θ act The actual steering angle fed back by the encoder.

[0083] Specifically, the edge computing node receives the calibrated steering resistance torque and the actual value from the torque sensor in real time, performing a difference calculation to determine the torque deviation. This deviation not only reflects the impact of soil condition changes on steering resistance, but may also include factors such as model prediction error and sensor noise.

[0084] The target steering angle is derived from the path planning algorithm and can be pre-set based on the tractor's operating task and terrain information. The actual steering angle is fed back in real time by an encoder installed in the steering system. The edge computing node receives these two angle values and performs a subtraction operation to determine the path tracking angle error. This error directly reflects the degree of deviation between the tractor's current steering state and the planned path.

[0085] S42, inputting the torque deviation and the path tracking angle error into the fuzzy inference model, and outputting the PID parameter correction based on the rule base optimized by the genetic algorithm; the correction formula of the PID parameter correction is:

[0086]

[0087] Where ΔK p is the correction value of the proportional coefficient in the PID parameter, ΔK i is the correction value of the integral coefficient in the PID parameter, ΔK dis the correction amount of the differential coefficient in the PID parameters, t is the time; α, β, δ, ε are the weight coefficients optimized by the genetic algorithm and determined by reinforcement learning training; sgn(x) is the sign function, when x>0, sgn(x)=1, when x=0, sgn(x)=0, when x<0, sgn(x)=-1.

[0088] Specifically, the core of the fuzzy inference model lies in the fuzzy rule base, which uses fuzzy logic to handle uncertainty and nonlinear relationships. The model receives torque deviation and path tracking angle error as input. The fuzzy rule base defines a series of fuzzy rules based on the magnitude and trend of the deviations, such as "If both the torque deviation and the path error are large, increase the proportional coefficient." These rules fuzzify the deviations, mapping them to fuzzy linguistic variables (such as "small," "medium," and "large"). Logical reasoning is then performed, and finally, defuzzification is used to determine the precise PID parameter corrections.

[0089] To improve the performance of the fuzzy inference model, the weight coefficients α, β, δ, and ε in the rule base are optimized using a genetic algorithm. This algorithm simulates the process of biological evolution, gradually optimizing the weight coefficients through screening, crossover, and mutation of the initial population. The goal of optimization is to ensure that the PID parameter corrections output by the model minimize path tracking error and torque deviation under various operating conditions. Specifically, the genetic algorithm uses path tracking accuracy and steering stability as fitness functions, and through multiple generations of evolution, it determines the optimal combination of weight coefficients.

[0090] PID parameter corrections are calculated based on the optimized rule base. The proportional coefficient correction is determined by the sign of the torque deviation and the absolute value of the path error, reflecting the immediate response to the current deviation. The integral coefficient correction considers the cumulative effect of the torque deviation over time and aims to eliminate the system's static error. The differential coefficient correction is based on the rate of change of the torque deviation and the actual steering angle speed, and is used to suppress system oscillations and improve dynamic response speed.

[0091] S43, updating the PID parameters based on the PID parameter correction amount to obtain the corrected PID parameters; the calculation formula of the corrected PID parameters is:

[0092]

[0093] Among them, K′ p is the corrected proportional coefficient, K′ i is the corrected integral coefficient, K′ d is the modified differential coefficient; K p is the proportionality coefficient, K i is the integral coefficient, K d is the differential coefficient.

[0094] Specifically, the updated PID parameter K′ p , K′ i , K′ d Used in tractor steering control systems. Proportional coefficient K' p Mainly affects the system's ability to quickly respond to the current deviation, the integral coefficient K' i Used to eliminate steady-state errors, and the differential coefficient K' d This helps suppress overshoot and oscillation in the system. By updating PID parameters in real time, the tractor's steering control system can dynamically adapt to changes in soil conditions and path tracking requirements, thereby achieving high-precision steering control.

[0095] In an optional embodiment, S6 includes the following steps:

[0096] S61. Upload the calibrated steering resistance torque, actual steering angle, and energy consumption of multiple tractors to the cloud to build a global data set.

[0097] Specifically, during each tractor's operation, the edge computing node collects this data in real time and periodically uploads it to the cloud server via wireless communication modules. After structured processing, the uploaded data forms a global dataset.

[0098] S62. Based on the global data set, the friction coefficient mapping table is updated using a distributed reinforcement learning algorithm. The optimization objective function of the distributed reinforcement learning algorithm is:

[0099]

[0100] Where μ is the friction coefficient, N is the number of data sets involved in the calculation, and each data set contains the calibrated steering resistance torque, actual steering angle, and energy consumption of the same tractor at the same time. T′ r,j is the calibrated steering resistance torque of the jth data set, T′ real,j is the actual steering angle of the jth data set, E j is the energy consumption of the jth data group, and η is the energy consumption weight coefficient.

[0101] Specifically, based on the global data set, a distributed reinforcement learning algorithm is used to update the friction coefficient mapping table. The optimization goal of the algorithm is to minimize the square error between the predicted steering resistance torque and the actual steering resistance torque of all data groups, while considering the impact of energy consumption.

[0102] The distributed reinforcement learning algorithm optimizes the friction coefficient mapping table through collaborative learning among multiple agents (corresponding to multiple tractors). The core idea of the algorithm is as follows:

[0103] 1) Multi-agent collaboration: Each tractor acts as an agent, generating data during operation and uploading it to the cloud. Agents do not communicate directly with each other, but collaborate indirectly by sharing a global dataset.

[0104] 2) Strategy Update: The cloud server regularly extracts data from the global dataset and uses an optimization algorithm to update the friction coefficient mapping table. The updated mapping table is then sent to each edge computing node via the wireless network.

[0105] 3) Reward Function Design: The optimization objective function comprehensively considers steering accuracy and energy consumption. Steering accuracy is measured by the error between the predicted steering resistance torque and the actual value, while energy consumption is used as a penalty term to ensure that steering performance is optimized while energy consumption is reduced.

[0106] 4) Strategy Evaluation: After receiving the new friction coefficient mapping table, the agent applies it to its tasks and uploads the new data to the cloud. The cloud evaluates the effectiveness of the updated strategy by comparing the performance metrics of the old and new data.

[0107] The aforementioned remote intelligent control method for new energy tractors obtains soil micromorphology data, stress parameter data, and moisture data, and generates dynamic soil cohesion and friction angle through Kalman filtering. A steering resistance torque model is constructed by querying a friction coefficient mapping table based on these parameters. The friction coefficient is calibrated in the cloud using the torque sensor's measured values to generate a resistance torque calibration value. The calibrated resistance torque and path tracking error are input into a fuzzy inference model to dynamically adjust the proportional, integral, and differential parameters of the PID controller. A target steering angle is generated based on the corrected PID parameters, and the steering motor is driven by a control algorithm. The tractor's state data during the steering process is collected. Finally, this state data is uploaded to the cloud, and the friction coefficient mapping table is optimized through reinforcement learning. This method enables real-time and accurate prediction of steering resistance and adaptive matching of control parameters in complex soil environments, significantly improving path tracking accuracy and system stability, reducing energy consumption, and ensuring the safety and reliability of remote control.

[0108] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0109] Based on the same inventive concept, the present application also provides a system for implementing the aforementioned remote new energy tractor intelligent control method. The solution provided by this system is similar to the solution described in the aforementioned method. Therefore, the specific limitations in the following embodiments of one or more remote new energy tractor intelligent control systems can be found in the aforementioned definitions of the remote new energy tractor intelligent control method and will not be further elaborated here.

[0110] In an exemplary embodiment, Figure 3 As shown, a remote new energy tractor intelligent control system 30 is provided, which is applied to the method in the above method embodiment. The system includes a data acquisition module 31, a control module 32, a cloud server 33 and a locally deployed edge computing node 34; the cloud server includes a friction coefficient query module 331 and a friction coefficient calibration module 332; the edge computing node includes a Kalman filter fusion module 341, a model construction module 342, a model update module 343, a PID parameter correction module 344 and a control parameter generation module 345;

[0111] The data acquisition module 31 is used to obtain soil microscopic morphology data, stress parameter data and moisture data;

[0112] A Kalman filter fusion module 341 is configured to receive the microscopic topography data, stress parameter data, and humidity data sent by the data acquisition module, and fuse the microscopic topography data, stress parameter data, and humidity data through Kalman filtering to obtain soil cohesion and friction angle;

[0113] The friction coefficient query module 331 is used to receive the soil cohesion and friction angle sent by the Kalman filter fusion module, and query the initial friction coefficient through a preset friction coefficient mapping table based on the soil cohesion and friction angle;

[0114] The model building module 342 is configured to receive the initial friction coefficient sent by the friction coefficient query module and the soil cohesion and friction angle sent by the Kalman filter fusion module; and to build a steering resistance torque model based on the soil cohesion, friction angle, and initial friction coefficient;

[0115] The friction coefficient calibration module 332 is configured to receive the measured operating parameters of the tractor uploaded by the data acquisition module, and calibrate the initial friction coefficient based on the measured operating parameters to obtain a calibrated friction coefficient;

[0116] The model updating module 343 is configured to receive the calibrated friction coefficient sent by the friction coefficient calibration module, and update the steering resistance torque model according to the calibrated friction coefficient to obtain the calibrated steering resistance torque;

[0117] a PID parameter correction module 344 for receiving the calibrated steering resistance torque sent by the steering resistance torque model update module, and correcting the PID parameters in real time using a fuzzy inference model based on the calibrated steering resistance torque and the path tracking error to obtain corrected PID parameters;

[0118] The control parameter generation module 345 is configured to receive the corrected PID parameters sent by the PID parameter correction module, generate a target steering angle based on the corrected PID parameters, and generate a control instruction based on the target steering angle through a control algorithm;

[0119] The control module 32 is used to receive the control instructions sent by the control parameter generation module, drive the steering motor to perform steering according to the control instructions, and trigger the data acquisition module to collect the status data of the tractor during the steering process;

[0120] The friction coefficient query module 331 is further configured to receive the status data uploaded by the data acquisition module and optimize the friction coefficient mapping table according to the status data using reinforcement learning.

[0121] An embodiment of the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0122] An embodiment of the present application further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0123] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0124] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A remote intelligent control method for new energy tractors, characterized in that: The method comprises: S1. Acquire soil micromorphology data, stress parameter data, and humidity data, and fuse the micromorphology data, stress parameter data, and humidity data through Kalman filtering to obtain soil cohesion and friction angle; S2. Based on the soil cohesion and the friction angle, querying an initial friction coefficient using a preset friction coefficient mapping table on the cloud; constructing a steering resistance torque model on a locally deployed edge computing node based on the soil cohesion, the friction angle, and the initial friction coefficient; S3. Calibrate the initial friction coefficient on the cloud based on the measured operating parameters of the tractor to obtain a calibrated friction coefficient, and update the steering resistance torque model of the edge computing node according to the calibrated friction coefficient to obtain a calibrated steering resistance torque; S4. Correcting PID parameters in real time using a fuzzy inference model based on the calibrated steering resistance torque and path tracking error to obtain corrected PID parameters; S5. Generate a target steering angle based on the corrected PID parameters; and drive the steering motor to perform steering according to the target steering angle through a control algorithm, and collect state data of the tractor during the steering process; S6. Feedback the state data to the cloud, and optimize the friction coefficient mapping table according to the state data using reinforcement learning.

2. The method according to claim 1, characterized in that Said S1 comprises: S11, scanning the surface topography by a laser radar, and extracting the local topographic slope of the soil as the microscopic topography data; S12. Measuring compressive stress and shear stress using a soil triaxial sensor as stress parameter data; S13, receiving infrared band data from the satellite, and calculating the soil moisture data using a normalized differential moisture index; S14. Perform Kalman filtering iteration on the local terrain slope, the compressive stress, the shear stress, and the humidity data to generate the dynamic soil cohesion and the friction angle; the expressions of the soil cohesion and the friction angle are: Where C is the soil cohesion, φ is the friction angle, σ s is the shear stress, is the maximum value of the shear stress, σ p is the compressive stress, ω soil is the humidity data, γ is the local terrain slope, k0 is the humidity correction factor, Δφ(γ,ω soil ) is the terrain-humidity compensation function.

3. The method according to claim 2, characterized in that The expression of the steering resistance torque model is: Among them, T r is the steering resistance torque, μ0 is the initial friction coefficient, R w is the wheel radius of the tractor, k s is the slip attenuation factor; S is the wheel slip rate, defined as v is the traveling speed of the tractor, and ω is the angular velocity of the wheels of the tractor.

4. The method according to claim 3, characterized in that The S3 includes: S31. Collecting the actual measured value of the torque sensor of the tractor, calculating the initial steering resistance torque using the steering resistance torque model, and uploading the initial steering resistance torque and the actual measured value of the torque sensor to an edge computing node; S32. On the edge computing node, compare the initial steering resistance torque with the actual value measured by the torque sensor. If the relative error is greater than a preset error threshold, trigger the cloud to calibrate the initial friction coefficient to obtain the calibrated friction coefficient, and send the calibrated friction coefficient to the edge computing node; wherein, the relative error is T real is the actual measured value of the torque sensor; S32. On the edge computing node, update the steering resistance torque model according to the calibrated friction coefficient to obtain the calibrated steering resistance torque; the expression of the updated steering resistance torque model is: Among them, T′ r is the steering resistance torque after calibration, and μ′ is the friction coefficient after calibration.

5. The method according to claim 4, characterized in that The triggering cloud to calibrate the initial friction coefficient to obtain the calibrated friction coefficient, and sending the calibrated friction coefficient to the edge computing node includes: S321. Establishing a dynamic digital twin model of the interaction between the tractor and the soil based on a physical engine; S322: Injecting the actual measured value of the torque sensor into the dynamic digital twin model, and iteratively adjusting the initial friction coefficient until the absolute value of the difference between the simulated steering resistance torque value in the simulation result output by the dynamic digital twin model and the actual measured value of the torque sensor is less than or equal to a preset value, thereby obtaining the calibrated friction coefficient; S323: encrypt the calibrated friction coefficient via a wireless network and transmit it to an edge computing node.

6. The method according to claim 5, characterized in that The S4 includes: S41, calculating the torque deviation according to the calibrated steering resistance torque and calculating the path tracking angle error; wherein the torque deviation is ΔT r =T′ r -T real , the path tracking angle error is e θ =θ cmd -θ act ,θ cmd is the target steering angle of the planned path, θ act The actual steering angle fed back by the encoder; S42, inputting the torque deviation and the path tracking angle error into the fuzzy inference model, and outputting a PID parameter correction based on a rule base optimized by a genetic algorithm; the correction formula of the PID parameter correction is: Where ΔK p is the correction value of the proportional coefficient in the PID parameter, ΔK i is the correction value of the integral coefficient in the PID parameter, ΔK d is the correction value of the differential coefficient in the PID parameters, t is time; α, β, δ, ε are weight coefficients optimized by genetic algorithm and determined by reinforcement learning training; sgn(x) is the sign function, when x>0, sgn(x)=1, when x=0, sgn(x)=0, when x<0, sgn(x)=-1; S43. Update the PID parameter based on the PID parameter correction amount to obtain the corrected PID parameter; the calculation formula of the corrected PID parameter is: Among them, K′ p is the corrected proportional coefficient, K′ i is the corrected integral coefficient, K′ d is the modified differential coefficient; K p is the proportional coefficient, K i is the integral coefficient, K d is the differential coefficient.

7. The method according to any one of claims 1 to 6, characterized in that The S6 includes: S61, uploading the calibrated steering resistance torque, the actual steering angle, and the energy consumption of multiple tractors to the cloud to construct a global data set; S62: Based on the global data set, update the friction coefficient mapping table using a distributed reinforcement learning algorithm; the optimization objective function of the distributed reinforcement learning algorithm is: Where μ is the friction coefficient, N is the number of data sets involved in the calculation, each of which contains the calibrated steering resistance torque, actual steering angle, and energy consumption of the same tractor at the same time, T′ ,j is the calibrated steering resistance torque of the jth data set, T′ real,j is the actual steering angle of the jth data set, E j is the energy consumption of the jth data group, and η is the energy consumption weight coefficient.

8. A remote new energy tractor intelligent control system, applied to the method according to any one of claims 1 to 7, characterized in that: The system includes a data acquisition module, a control module, a cloud server, and a locally deployed edge computing node; the cloud server includes a friction coefficient query module and a friction coefficient calibration module; the edge computing node includes a Kalman filter fusion module, a model construction module, a model update module, a PID parameter correction module, and a control parameter generation module; The data acquisition module is used to acquire soil microscopic morphology data, stress parameter data and humidity data; The Kalman filter fusion module is used to receive the micro-morphology data, the stress parameter data, and the humidity data sent by the data acquisition module, and fuse the micro-morphology data, the stress parameter data, and the humidity data through Kalman filtering to obtain soil cohesion and friction angle; The friction coefficient query module is configured to receive the soil cohesion and the friction angle sent by the Kalman filter fusion module, and query an initial friction coefficient through a preset friction coefficient mapping table based on the soil cohesion and the friction angle; The model building module is configured to receive the initial friction coefficient sent by the friction coefficient query module and the soil cohesion and the friction angle sent by the Kalman filter fusion module; and to build a steering resistance torque model based on the soil cohesion, the friction angle, and the initial friction coefficient; The friction coefficient calibration module is configured to receive the measured operating parameters of the tractor uploaded by the data acquisition module, and calibrate the initial friction coefficient based on the measured operating parameters to obtain a calibrated friction coefficient; The model updating module is configured to receive the calibrated friction coefficient issued by the friction coefficient calibration module, and update the steering resistance torque model according to the calibrated friction coefficient to obtain the calibrated steering resistance torque; The PID parameter correction module is configured to receive the calibrated steering resistance torque sent by the steering resistance torque model update module, and to correct the PID parameters in real time using a fuzzy inference model according to the calibrated steering resistance torque and a path tracking error to obtain corrected PID parameters; The control parameter generating module is configured to receive the corrected PID parameters sent by the PID parameter correcting module and generate a target steering angle based on the corrected PID parameters; generating a control instruction through a control algorithm according to the target steering angle; The control module is configured to receive the control instruction sent by the control parameter generation module, drive the steering motor to perform steering according to the control instruction, and trigger the data acquisition module to collect status data of the tractor during the steering process; The friction coefficient query module is further configured to receive the status data uploaded by the data acquisition module, and optimize the friction coefficient mapping table according to the status data using reinforcement learning.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images