Agricultural machinery intelligent navigation system based on fixed time generalized super-distortion control

Abstract

The invention discloses an agricultural machinery intelligent navigation system based on fixed time generalized super-distortion control, which combines a Beidou satellite original signal, an agricultural machinery dynamical model, an improved A * optimization algorithm and a sliding mode surface function to form a closed loop structure, correspondingly outputs an optimal control law of an optimal path to perform agricultural machinery interaction, and realizes high stability, high precision and high reliability. And high-precision and high-dynamic obstacle avoidance capability agricultural operation is realized. The system is suitable for various agricultural machines such as an electric tractor and a seeding machine, and aims to realize centimeter-level positioning precision, path tracking control of fixed time convergence, signal transmission delay reduction, path tracking error reduction and stable communication in a global range. And a more accurate and advanced solution is provided for modern agricultural operation.

Description

Technical Field

[0001] The present invention relates to the field of agricultural machinery and equipment, and particularly to an intelligent navigation system for agricultural machinery based on fixed-time generalized super-twisting control. Background Art

[0002] With the development of precision agriculture technology, the intelligent navigation system for agricultural machinery plays a key role in improving operation efficiency and reducing labor intensity. Traditional navigation systems adopt algorithms such as PID control or classical sliding mode control. Limited by complex farmland environment interferences such as multipath effects and signal occlusion, the positioning accuracy often fails to meet the millimeter-level deviation correction requirements for inter-ridge operations. The traditional sliding mode control algorithm has chattering problems, the convergence time depends on the initial state, and the feedback delay is relatively high, resulting in limited dynamic path tracking performance. Therefore, it is easy to cause problems such as control delay and insufficient stability in the control system. The GPS positioning technology is affected by multipath effects and ionospheric interference, with large positioning errors and difficult to meet the requirements of precision agriculture. The static obstacle avoidance algorithm cannot adapt to the dynamic obstacle environment commonly existing in farmland, resulting in a further increase in the path error reporting rate, and thus the robustness is affected.

[0003] In the prior art, improvement schemes such as RTK-GPS positioning, etc., although partially improve the performance but still limited. Therefore, there is an urgent need to construct an integrated navigation system that integrates high-precision environment perception, agricultural machinery motion characteristic constraints and strong robust control to achieve collaborative control of precise trajectory tracking and dynamic obstacle avoidance in complex farmland scenarios. Summary of the Invention

[0004] Object of the Invention: To solve the problems mentioned in the background art, the present invention discloses an intelligent navigation system for agricultural machinery based on fixed-time generalized super-twisting control. By combining the original Beidou satellite signal, the agricultural machinery dynamics model, the improved A* optimization algorithm and the sliding mode surface function to form a closed-loop structure, the optimal control law corresponding to the optimal path is output for agricultural machinery interaction, so as to achieve agricultural operations with high stability, high precision and high dynamic obstacle avoidance capabilities.

[0005] Technical Solution:

[0006] The present invention discloses an intelligent navigation system for agricultural machinery based on fixed-time generalized super-twisting control. The system includes a Beidou navigation and positioning module, a path optimization module and a generalized super-twisting control module;

[0007] The Beidou navigation and positioning module is used to receive the original positioning data of Beidou navigation satellites, obtain the high-precision pose information of the current agricultural machinery after preprocessing, and combine the electronic fence status identifier and the obstacle distribution heat map to generate a risk field modeling by using the RTK double-difference ambiguity fixing technology;

[0008] The path optimization module is connected to the Beidou navigation and positioning module, and is used to receive the data stream output by the Beidou navigation and positioning module. By building an agricultural machinery dynamics model, based on the high-precision pose information of the agricultural machinery, the curvature and smoothness of the trajectory generation are constrained, and the B-spline curve interpolation algorithm is combined to optimize the curvature of the input path to make the path smooth; the cost function of the improved A* algorithm is used to optimize the path to achieve dynamic obstacle avoidance, and finally the planned path is output.

[0009] The generalized super-twisting control module is connected to the path optimization module, and a delay compensation unit and PSO gain optimization are deployed in nested mode. It is connected in parallel with the sliding mode observer to form a feedforward-feedback composite controller, which is used to receive the planned path, adopt a sliding mode control strategy with fixed-time convergence to optimize the planning error to obtain the final path, and output the optimal control law based on the final path.

[0010] Further, the preprocessing is as follows:

[0011] The original data is transmitted back by Beidou navigation satellites, the positioning error is eliminated by combining RTK differential correction technology, and it is fused with the data obtained by IMU inertial measurement technology to calculate the high-precision pose information of the agricultural machinery.

[0012] The risk field modeling process is as follows:

[0013] High-precision signal receivers are arranged, and the RTK double-difference ambiguity fixing technology is used to eliminate the ionospheric error, and the received signal covariance matrix R = E[yy H is constructed, where E is the expected value operator, y is the received signal vector, and y H is the conjugate device of y.

[0014] The direct signal component is extracted through eigenvalue decomposition, and the positioning solution is optimized based on the weighted least squares method. The objective function is:

[0015] min x ||W(z - Hx)|| 2

[0016] where W is the weight matrix, z is the observed value, H is the design matrix, and the multipath suppression unit: the distance d_i from the agricultural machinery to the i-th obstacle is obtained through the Beidou signal, and the risk field modeling function is constructed:

[0017] R(x, y) = Σ(wi·e^(-di^2 / (2σ^2)))

[0018] where wi is the weight coefficient of the i-th obstacle, di is the Euclidean distance from the agricultural machinery to the i-th obstacle, and σ is the standard deviation of the Gaussian distribution, which controls the attenuation rate of the risk field.

[0019] Further, the specific implementation of the path optimization module to optimize the curvature of the input path is as follows:

[0020] Based on the agricultural machinery pose information of the Beidou navigation and positioning module, combined with the farmland layer obtained by GIS, the geometric boundary of the farmland where it is located and the distribution of soil properties are obtained. By inputting the agricultural machinery pose information, IMU inertial measurement data, risk field modeling, as well as the initial control instructions of the agricultural machinery and the surrounding environment parameters obtained by sensors, an agricultural machinery dynamics model is built; data preprocessing is carried out, and the input includes: Beidou coordinates (x, y), IMU data (a x , a y , ω z ), the initial path instruction (x ref , y ref ), and environmental parameters (μ, φ). The positioning and inertial data are fused through the extended Kalman filter (EKF) to eliminate noise and estimate the real-time pose (x, y, θ, v) of the agricultural machinery;

[0021] The state equation of the agricultural machinery dynamics model is as follows:

[0022]

[0023] The observation equation is as follows:

[0024]

[0025] Among them, x k represents the state vector at the kth moment, ω z is the angular velocity measured by the IMU, and w k is the process noise vector, which follows a Gaussian distribution.

[0026] Dynamic constraints are carried out: in the steering dynamic constraints, according to the wheelbase L and the maximum steering angle δ max of the agricultural machinery parameters, the minimum turning radius is calculated to constrain the path curvature to avoid mechanical rollover caused by sharp turns; the maximum driving force F max = μ·m·gcosφ, and the acceleration is restricted. Among them, the rolling resistance F roll = C r ·m·g; slope compensation is introduced, and the gravity component F g = m·g·sinφ. The driving force balance equation is introduced to correct the acceleration limit. Combining the built dynamics model, the B-spline curve interpolation algorithm is used to optimize the path curvature. The control point update strategy is: Q i = P i + λ·(P i+1 - P i+1 ), ensuring compliance with the kinematic constraints of the agricultural machinery;

[0027] Furthermore, the specific implementation of the improved A* algorithm is as follows:

[0028] The traditional cost function of the A* algorithm is as follows:

[0029] f(n) = g(n) + h(n)

[0030] where g(n) is the actual cost from the starting point to the current node n, and h(n) is the estimated cost from the current node to the end point;

[0031] The cost function of the improved A* algorithm is as follows:

[0032]

[0033] Compared with the original A* algorithm, the sensitivity of dynamic obstacle speed direction adjustment for dynamic avoidance is newly added, making the path far away from the obstacles moving at high speed:

[0034]

[0035] where, represents the dynamic obstacle speed direction vector near node n; γ is the weight coefficient;

[0036] The environmental uncertainty entropy value is newly added to quantify the environmental uncertainty and improve the dynamic obstacle avoidance ability:

[0037] Entropy(P obs ) = -∑P obs log P obs

[0038] where, Entropy(P obs ) represents the uncertainty entropy value, and P obs is a dynamic parameter.

[0039] Furthermore, the generalized super-twisting control module adopts a fixed-time convergence sliding mode control strategy, and the specific operation is as follows:

[0040] The sliding mode surface function is defined as:

[0041]

[0042] where, e(t) = x desired (t) - x actual (t) is the tracking error, and k is the integral gain;

[0043] The control law is:

[0044]

[0045] where α, β, γ are control parameters used to ensure the convergence of the system within a fixed time.

[0046] Through the Lyapunov function Verify the stability to ensure that the system is in a fixed time Deploy the introduced delay compensation unit and PSO gain optimization in the generalized super-twisting control module, control the timing reconstruction, and form a feedforward-feedback composite controller in parallel with the sliding mode observer.

[0047] Furthermore, the feedforward-feedback composite controller is implemented as follows:

[0048] Establish the system delay transfer function through the step response experiment Quantify the cumulative delay τ of links such as signal transmission and hydraulic response;

[0049] Based on the delay inverse model Generate the predictive control quantity u ff =K p ·r(t + τ), and inject it into the actuator in advance;

[0050] Use the super-twisting algorithm to calculate the error correction quantity:

[0051] u fb =λ|e(t)| 1 / 2 sign(e(t)) + α∫sign(e(t))dt

[0052] Suppress the residual deviation;

[0053] Define the fitness function:

[0054]

[0055] Balance the tracking accuracy and energy loss;

[0056] Initialize the particle swarm, and the optimized variables are the feedforward gain K p And the sliding mode parameters λ, α. After iteration, obtain the Pareto optimal solution and embed the extended state observer ESO disturbance feedforward compensation.

[0057] Furthermore, the disturbance modeling of the extended state observer ESO disturbance feedforward compensation is as follows:

[0058] Regard the external disturbance d(t) and the model uncertainty Δf as the extended state x3, and construct a third-order ESO:

[0059]

[0060] Among them, β1, β2, β3 are configured according to the bandwidth method as ω0 = 25rad / s corresponding to the typical disturbance frequency band of farmland;

[0061] Feedforward compensation: The estimated total disturbance Injection control law:

[0062]

[0063] where u GSMC is the output of the original super-twisting controller, is the total disturbance output.

[0064] Dynamic parameter adjustment: When the IMU detects that the agricultural machine enters the muddy area, ω0 is automatically increased to 40 rad / s to improve the high-frequency disturbance capture ability.

[0065] Beneficial effects:

[0066] 1. Relying on the Beidou high-precision positioning technology, this invention preprocesses the original satellite data by combining RTK and IMU, calculates the high-precision pose information of the agricultural machine (<2 cm), improves the multipath suppression ability by 60%, effectively overcomes the problem of signal interference caused by obstacles such as buildings and trees in the farmland environment, and ensures the operation consistency.

[0067] 2. By combining the agricultural machine dynamics model with the improved A* algorithm for dynamic obstacle avoidance, this invention further optimizes the path curvature and dynamic obstacle avoidance of the agricultural machine navigation route, further improves the robustness of the navigation system, reduces path errors, and at the same time reduces the accident rate of the agricultural machine and the maintenance cost, thereby reducing the cost.

[0068] 3. By deploying the delay compensation unit and PSO gain optimization, and forming a feedforward-feedback composite controller in parallel with the sliding mode observer, this invention adopts the sliding mode control strategy with fixed-time convergence to optimize the planning error to obtain the final path, further reduces the error, and improves the response speed and operation accuracy of the machine. Description of the drawings

[0069] Figure 1 is the system operation flowchart of this invention;

[0070] Figure 2 is the schematic diagram of the generalized super-twisting control module of this invention. Detailed implementation manners

[0071] To make the objectives, technical solutions and advantages of the embodiments of this invention clearer, the technical solutions in the embodiments of this invention will be clearly and completely described below with reference to the drawings in the embodiments of this invention. Apparently, the described embodiments are some but not all of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative efforts shall fall within the protection scope of this invention.

[0072] As Figure 1-2As shown in the figure, the present invention discloses an intelligent navigation system for agricultural machinery based on fixed-time generalized super-twisting control. The system includes a Beidou navigation and positioning module, a path optimization module, and a generalized super-twisting control module;

[0073] The Beidou navigation and positioning module is used to receive the original positioning data of Beidou navigation satellites, obtain the high-precision pose information of the current agricultural machinery after preprocessing, combine the status identification of the electronic fence and the heat map of obstacle distribution, and generate a risk field model using the RTK double-difference ambiguity fixing technology.

[0074] The preprocessing is as follows:

[0075] The original data is transmitted back by Beidou navigation satellites, the positioning error is eliminated by combining the RTK differential correction technology, and the data obtained by the IMU inertial measurement technology is fused to calculate the high-precision pose information of the agricultural machinery;

[0076] The process of risk field modeling is as follows:

[0077] High-precision signal receivers are arranged, the ionospheric error is eliminated using the RTK double-difference ambiguity fixing technology, and the covariance matrix R of the received signal is constructed as R = E[yy H where E is the expected value operator, y is the received signal vector, and y H is the conjugate device of y.

[0078] The direct signal component is extracted through eigenvalue decomposition, and the positioning solution is optimized based on the weighted least squares method. The objective function is:

[0079] min x ||W(z - Hx)|| 2

[0080] where W is the weight matrix, z is the observed value, H is the design matrix, and the multipath suppression unit: the distance d_i from the agricultural machinery to the i-th obstacle is obtained through the Beidou signal, and the risk field modeling function is constructed as:

[0081] R(x,y) = Σ(w_i·e^(-d_i^2 / (2σ^2)))

[0082] where w_i is the weight coefficient of the i-th obstacle, d_i is the Euclidean distance from the agricultural machinery to the i-th obstacle, and σ is the standard deviation of the Gaussian distribution, which controls the attenuation rate of the risk field.

[0083] The path optimization module is connected to the Beidou navigation and positioning module, and is used to receive the data stream output by the Beidou navigation and positioning module. Through the built-in agricultural machinery dynamics model, based on the high-precision pose information of the agricultural machinery, the curvature and smoothness of the generated trajectory are constrained, and the input path curvature is optimized by combining the B-spline curve interpolation algorithm to make the path smooth; the path is optimized by the cost function of the improved A* algorithm to achieve dynamic obstacle avoidance, and finally the planned path is output;

[0084] The path optimization module optimizes the curvature of the input path as follows:

[0085] Based on the agricultural machinery pose information from the Beidou navigation and positioning module, combined with the farmland layer obtained from GIS, the geometric boundaries of the farmland and the distribution of soil properties are obtained. By inputting the agricultural machinery pose information, IMU inertial measurement data, risk field modeling, initial control instructions for the agricultural machinery, and ambient parameters obtained by sensors, an agricultural machinery dynamics model is built; data preprocessing is carried out, and the input includes: Beidou coordinates (x, y), IMU data (a x , a y , ω z ), initial path instructions (x ref , y ref ), and ambient parameters (μ, φ). The positioning and inertial data are fused through the extended Kalman filter (EKF) to eliminate noise and estimate the real-time pose (x, y, θ, v) of the agricultural machinery;

[0086] The state equation of the agricultural machinery dynamics model is as follows:

[0087]

[0088] The observation equation is as follows:

[0089]

[0090] Among them, x k represents the state vector at the k-th moment, ω z is the angular velocity measured by the IMU, w k is the process noise vector, which follows a Gaussian distribution.

[0091] Dynamics constraints are carried out: in the steering dynamics constraints, according to the wheelbase L and the maximum steering angle δ max of the agricultural machinery parameters, the minimum turning radius is calculated to constrain the path curvature to avoid mechanical rollover caused by sharp turns; the maximum driving force F max = μ·m·gcosφ, which restricts the acceleration Among them, the rolling resistance F roll = C r ·m·g; slope compensation is introduced, and the gravity component F g = m·g·sinφ. The driving force balance equation is introduced to correct the acceleration limit. Combining the built dynamics model, the B-spline curve interpolation algorithm is used to optimize the path curvature, and the control point update strategy is: Q i = P i + λ·(P i+1 - P i+1 ), ensuring compliance with the kinematic constraints of the agricultural machinery;

[0092] The specific implementation of the improved A* algorithm is as follows:

[0093] The traditional cost function of the A* algorithm is:

[0094] f(n) = g(n) + h(n)

[0095] where g(n) is the actual cost from the starting point to the current node n, and h(n) is the estimated cost from the current node to the end point;

[0096] The cost function of the improved A* algorithm is:

[0097]

[0098] Compared with the original A* algorithm, the sensitivity of dynamic obstacle speed direction adjustment for dynamic avoidance is newly added, making the path far away from high-speed moving obstacles:

[0099]

[0100] where, represents the dynamic obstacle speed direction vector near node n; γ is the weight coefficient;

[0101] The newly added environmental uncertainty entropy value quantifies the environmental uncertainty and improves the dynamic obstacle avoidance ability:

[0102] Entropy(P obs ) = -∑P obs log P obs

[0103] where, Entropy(P obs ) represents the uncertainty entropy value, and P obs is a dynamic parameter.

[0104] The generalized super-twisting control module is connected to the path optimization module, with a delay compensation unit and PSO gain optimization nested and deployed inside, and is connected in parallel with the sliding mode observer to form a feedforward-feedback composite controller, which is used to receive the planned path, adopt a sliding mode control strategy with fixed-time convergence to optimize the planning error to obtain the final path, and output the optimal control law based on the final path.

[0105] The generalized super-twisting control module adopts a sliding mode control strategy with fixed-time convergence, and the specific operation is as follows:

[0106] The sliding mode surface function is defined as:

[0107]

[0108] where, e(t) = x desired (t) - x actual(t) is the tracking error and k is the integral gain;

[0109] The control law is:

[0110]

[0111] where α, β, and γ are control parameters used to ensure the convergence of the system within a fixed time.

[0112] Verify the stability through the Lyapunov function to ensure the system converges within a fixed time Deploy a delay compensation unit and PSO gain optimization nested within the generalized super-twisting control module, reconstruct the control timing, and form a feedforward-feedback composite controller in parallel with the sliding mode observer.

[0113] The feedforward-feedback composite controller is implemented as follows:

[0114] Establish the system delay transfer function through a step response experiment Quantify the cumulative delay τ in links such as signal transmission and hydraulic response;

[0115] Based on the delay inverse model Generate the predictive control quantity u ff = K p ·r(t + τ), and inject it into the actuator in advance;

[0116] Calculate the error correction quantity using the super-twisting algorithm:

[0117] u fb = λ|e(t)| 1 / 2 sign(e(t)) + α∫sign(e(t))dt

[0118] Suppress the residual deviation;

[0119] Define the fitness function:

[0120]

[0121] Balance the tracking accuracy and energy loss;

[0122] Initialize the particle swarm, and the optimization variable is the feedforward gain K p and the sliding mode parameters λ and α. After iteration, obtain the Pareto optimal solution and embed the extended state observer ESO disturbance feedforward compensation.

[0123] The disturbance modeling of the extended state observer ESO disturbance feedforward compensation is as follows:

[0124] Regard the external disturbance d(t) and the model uncertainty Δf as the extended state x3, and construct a third-order ESO:

[0125]

[0126] where β1, β2, and β3 are configured according to the bandwidth method as ω0 = 25 rad / s corresponds to the typical disturbance frequency band of farmland;

[0127] Feedforward compensation: Inject the estimated total disturbance into the control law:

[0128]

[0129] where u GSMC is the output of the original super-twisting controller, and is the total disturbance output.

[0130] Dynamic parameter adjustment: When the IMU detects that the agricultural machinery enters the muddy area, automatically increase ω0 to 40 rad / s to enhance the high-frequency disturbance capture ability.

[0131] The present invention realizes intelligent navigation through a closed-loop architecture of perception → planning → control. Through the modeling of the agricultural machinery pose information and the risk field obtained from step 1 (Beidou navigation and positioning module), combined with the optimized path obtained from the agricultural machinery dynamics model in step 2 (path optimization module) and the dynamic obstacle avoidance function obtained from the improved A* algorithm, finally, in step 3 (generalized super-twisting control scheme), the error is reduced by relying on the delay compensation unit and PSO gain optimization, and finally a control algorithm scheme with centimeter-level and high stability for agricultural machinery is realized.

[0132] In this embodiment, the device parameters, system parameters, and test results are as follows:

[0133] Agricultural machinery model: Electric tractor (rated power 80 kW, wheelbase L = 2.1 m, maximum steering angle δ_max = 35°)

[0134] Positioning module: Beidou high-precision positioning terminal (dual-frequency reception, RTK differential correction)

[0135] Sensors: IMU (triaxial accelerometer ±5g, gyroscope ±300° / s), lidar (detection range 30 m)

[0136] Communication protocol: LoRa + 4G dual-mode, bandwidth utilization optimized to 85%

[0137] Control parameters: FTGSTA sliding mode gains α = 1.2, β = 0.8, γ = 0.3, ESO observation bandwidth ω0 = 25 rad / s.

[0138] High-precision positioning and electronic fence generation:

[0139] Positioning data fusion:

[0140] The original positioning error of Beidou is ±10 cm. After RTK correction and IMU data fusion, the pose accuracy is improved to ±1.8 cm (measured standard deviation).

[0141] Multipath suppression: The polarization filtering algorithm eliminates the interference of reflected signals from irrigation facilities, and the horizontal positioning error is reduced by 62% (from ±5.3 cm to ±2.0 cm).

[0142] Electronic fence and obstacle heat map:

[0143] Based on Beidou clustering analysis, a dynamic obstacle heat map is generated, marking 3 high-risk areas (the moving speed of fertilizer bags ≤ 0.5 m / s).

[0144] The boundary error of the electronic fence ≤ 3 cm, and the over-limit alarm response time < 0.5 s.

[0145] Dynamic path planning and dynamic obstacle avoidance (path optimization module):

[0146] Path optimization:

[0147] The B-spline interpolation algorithm optimizes the path curvature, and the maximum curvature κ_max = 0.15 m-1 (theoretical minimum turning radius R_min = 1.4 m) to avoid sharp turns.

[0148] Improved A* algorithm for dynamic obstacle avoidance test:

[0149] The failure rate of traditional A* obstacle avoidance is 22%, and it is reduced to 4% after improvement (new speed direction weight γ = 0.5, entropy value weight η = 0.3).

[0150] The path length only increases by 8%, but the safety is significantly improved.

[0151] Slope compensation:

[0152] On an 8° slope, the driving force error is < 5% after gravity component compensation, and the acceleration limit a_max = 1.8 m / s 2 (2.2 m / s without compensation 2 , easy to slip).

[0153] Fixed-time control and delay compensation (generalized super-twisting control module)

[0154] Path tracking performance:

[0155] Under FTGSTA control, the median of the path tracking error approaches 0.3 cm (4.5 cm for PID control), and the overshoot < 1%.

[0156] The fixed convergence time T = 1.8 s (initial error e0 = 20 cm), meeting the Lyapunov stability condition.

[0157] Delay compensation effect:

[0158] The cumulative delay of signal transmission and hydraulic response τ = 120 ms. After feedforward-feedback compensation, the actual control delay is reduced to 60 ms (the traditional scheme is 240 ms).

[0159] After PSO optimization, the control energy consumption is reduced by 25% (fitness function J = 0.85 → 0.64).

[0160] Communication and system stability

[0161] Data packet loss rate: The packet loss rate in complex terrain is 3.2% (the traditional module is 18%), and the instruction transmission success rate reaches 96.8%.

[0162] Fault warning: The system detects motor overheating 10 minutes in advance (abnormal vibration frequency of IMU > 40 Hz), avoiding shutdown losses.

[0163] The above description of the embodiments enables those skilled in the art to implement or use the present invention. Various modifications to the embodiments will be obvious to those skilled in the art. The general principles of the present invention can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention should not be limited to the embodiments shown herein, but should cover the widest scope consistent with the principles and novel features disclosed in the present invention.

Claims

1. An intelligent navigation system for agricultural machinery based on fixed-time generalized super-twisting control, characterized in that, The system includes a Beidou navigation and positioning module, a path optimization module, and a generalized super-twisting control module; The Beidou navigation and positioning module is used to receive the original positioning data of Beidou navigation satellites, obtain the high-precision pose information of the current agricultural machinery after preprocessing, combine the status identification of the electronic fence and the heat map of obstacle distribution, and use the RTK double-difference ambiguity fixing technology to generate a risk field model; The path optimization module is connected to the Beidou navigation and positioning module, and is used to receive the data stream output by the Beidou navigation and positioning module. By means of the built-in agricultural machinery dynamics model, based on the high-precision pose information of the agricultural machinery, the curvature and smoothness of the generated trajectory are constrained, and the B-spline curve interpolation algorithm is combined to optimize the input path curvature to make the path smooth; the cost function of the improved A* algorithm is used to optimize the path to achieve dynamic obstacle avoidance, and finally the planned path is output; The generalized super-twisting control module is connected to the path optimization module, and a delay compensation unit and PSO gain optimization are deployed in nested form, and are connected in parallel with a sliding mode observer to form a feedforward-feedback composite controller, which is used to receive the planned path, adopt a sliding mode control strategy with fixed-time convergence to optimize the planning error to obtain the final path, and output the optimal control law based on the final path.

2. The intelligent navigation system for agricultural machinery based on fixed-time generalized super-twisting control according to claim 1, wherein The preprocessing is as follows: The original data is transmitted back by Beidou navigation satellites, the positioning error is eliminated by combining RTK differential correction technology, and the data obtained by IMU inertial measurement technology is fused to calculate the high-precision pose information of the agricultural machinery; The process of risk field modeling is as follows: Arrange high-precision signal receivers, adopt RTK double-difference ambiguity fixing technology to eliminate ionospheric errors, and construct the received signal covariance matrix R = E[yy H , where E is the expected value operator, y is the received signal vector, and y H is the conjugate device of y. Extract the direct signal component through eigenvalue decomposition, optimize the positioning solution based on the weighted least squares method, and the objective function is: min x ||W(z - Hx)|| 2 Among them, W is the weight matrix, z is the observation value, H is the design matrix, multi-path suppression unit: the distance d_i from the agricultural machinery to the i-th obstacle is obtained through Beidou signals, and a risk field modeling function is constructed: R(x,y) = Σ(w_i·e^(-d_i^2 / (2σ^2))) Among them, w_i is the weight coefficient of the i-th obstacle, d_i is the Euclidean distance from the agricultural machinery to the i-th obstacle, and σ is the standard deviation of the Gaussian distribution, which controls the attenuation rate of the risk field.

3. The intelligent navigation system for agricultural machinery based on fixed-time generalized super-twisting control according to claim 2, wherein The specific implementation of the path optimization module to optimize the input path curvature is as follows: Based on the agricultural machinery pose information of the Beidou navigation and positioning module, combined with the farmland layer obtained by GIS, the geometric boundary of the farmland and the distribution of soil properties are obtained. By inputting the agricultural machinery pose information, IMU inertial measurement data, risk field modeling, as well as the initial control instructions of the agricultural machinery and the surrounding environment parameters obtained by sensors, an agricultural machinery dynamics model is built; data preprocessing is carried out, and the inputs are: Beidou coordinates (x, y), IMU data (a x , a y , ω z ), initial path instructions (x ref , y ref ), environmental parameters (μ, φ). The positioning and inertial data are fused through the Extended Kalman Filter (EKF) to eliminate noise and estimate the real-time pose (x, y, θ, v) of the agricultural machinery; The state equation of the agricultural machinery dynamics model is as follows: The observation equation is as follows: where, x k represents the state vector at the k-th moment, ω z is the angular velocity measured by the IMU, w k is the process noise vector, which follows a Gaussian distribution; Apply dynamic constraints: In the steering dynamic constraints, based on the agricultural machinery parameters of the wheelbase L and the maximum steering angle δ max , calculate the minimum turning radius Constrain the path curvature To avoid mechanical rollover caused by sharp turns; the maximum driving force F max = μ·m·gcosφ, limit the acceleration where the rolling resistance F roll = C r ·m·g; introduce slope compensation, the gravity component F g = m·g·sinφ, introduce the driving force balance equation to correct the acceleration limit, and combine the established dynamic model to optimize the path curvature using the B-spline curve interpolation algorithm. The control point update strategy is: Q i = P i +λ·(P i+1 -P i+1 ), to ensure compliance with the kinematic constraints of the agricultural machinery.

4. The intelligent navigation system for agricultural machinery based on fixed-time generalized super-twisting control according to claim 3, wherein, The specific implementation of the improved A* algorithm is as follows: The traditional cost function of the A* algorithm is: f(n) = g(n) + h(n) Among them, g(n) is the actual cost from the starting point to the current node n, and h(n) is the estimated cost from the current node to the end point; The cost function of the improved A* algorithm is: Compared with the original A* algorithm, the sensitivity of dynamic obstacle avoidance is increased by adding the dynamic obstacle speed direction to adjust the dynamic avoidance, so that the path is far away from the fast-moving obstacles: Among them, represents the velocity direction vector of the dynamic obstacle near node n; γ is the weight coefficient; The environmental uncertainty entropy value is added to quantify the environmental uncertainty and improve the dynamic obstacle avoidance ability: Entropy(P obs ) = -∑P obs logP obs Among them, Entropy(P obs ) represents the uncertainty entropy value, and P obs is a dynamic parameter.

5. The intelligent navigation system for agricultural machinery based on fixed-time generalized super-twisting control according to claim 1 or 4, characterized in that, The generalized super-twisting control module adopts a sliding mode control strategy with fixed-time convergence, and the specific operation is as follows: The sliding mode surface function is defined as: where, e(t) = x desired (t) - x actual (t) is the tracking error, and k is the integral gain; The control law is: Among them, α, β, γ are control parameters, which are used to ensure the convergence of the system within a fixed time; Verify stability through the Lyapunov function to ensure that the system is in a fixed time Deploy the introduced delay compensation unit and PSO gain optimization nested within the generalized super-twisting control module to control the timing reconstruction, and form a feedforward-feedback composite controller in parallel with the sliding mode observer.

6. The intelligent navigation system for agricultural machinery based on fixed-time generalized super-twisting control according to claim 5, characterized in that, The implementation of the feedforward-feedback composite controller is as follows: Establish the system delay transfer function through a step response experiment Quantify the cumulative delay τ of links such as quantized signal transmission and hydraulic response; Based on the time-delay inverse model Generate the predictive control quantity u ff = K p ·r(t + τ), and inject it into the actuator in advance; The error correction amount is calculated by using the super-twisting algorithm: u fb = λ|e(t)| 1 / 2 sign(e(t)) + α∫sign(e(t))dt Suppress the residual deviation; Define the fitness function: Balance the tracking accuracy and energy loss; Initialize the particle swarm, and the optimization variable is the feedforward gain K p Obtain the Pareto optimal solution after iteration with the sliding mode parameters λ and α, and embed the extended state observer ESO disturbance feedforward compensation.

7. The intelligent navigation system for agricultural machinery based on fixed-time generalized super-twisting control according to claim 6, characterized in that, The disturbance modeling of the extended state observer ESO disturbance feedforward compensation is as follows: The external disturbance d(t) and the model uncertainty Δf are unified as the extended state x3, and a third-order ESO is constructed: where β1, β2, β3 are configured according to the bandwidth method as ω0 = 25 rad / s corresponds to the typical disturbance frequency band of farmland; Feedforward compensation: Inject the estimated total disturbance into the control law: Among them, u GSMC is the output of the original super-twisting controller, and is the total disturbance output; Dynamic parameter tuning: When the IMU detects that the agricultural machinery enters the muddy area, automatically increase ω0 to 40 rad / s to enhance the high-frequency disturbance capture ability.

Images