A method for intelligent control of header height in combine harvesters

Abstract

This invention discloses an intelligent control method for the header height of a combine harvester, comprising the following steps: S1: Establishing a geometric model of header motion and determining the functional relationship between the header's ground clearance and the adjustment amount of the header cylinder; S2: Using a multi-sensor data fusion method based on a BP neural network, preprocessing and performing neural network calculations on the data collected by the angle sensor to obtain the real-time ground clearance of the header; S3: Constructing a hierarchical control framework for header height based on a BP neural network fuzzy PID algorithm, performing fuzzy processing on the height deviation and the rate of change of deviation, adaptively adjusting the PID control parameters, and realizing real-time closed-loop control of the header height. This intelligent control method for the header height of a combine harvester improves the height detection accuracy through multi-sensor data fusion and optimizes the control response characteristics by combining a neural network fuzzy PID algorithm, thereby achieving real-time and precise control of the header height, reducing harvest losses and improving operational efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for agricultural machinery, specifically to an intelligent control method for the header height of a combine harvester. Background Technology

[0002] With the rapid development of agricultural modernization and intelligent agricultural machinery, the operational precision and intelligence level of combine harvesters have become core indicators for improving harvesting efficiency and reducing harvesting losses. In mechanized harvesting of low-pod-setting crops such as soybeans, header loss accounts for more than 80% of the total harvesting loss. The main causes of problems such as inconsistent stubble height, missed cuts, and soil erosion are insufficient precision in header height adjustment and delayed response.

[0003] The current domestic combine harvester header height control technology has obvious defects: (1) Traditional control relies on manual adjustment by the driver, which is greatly affected by experience and terrain, has a slow response and poor stubble consistency, and is prone to missed cutting or soil erosion by the cutter in complex undulating terrain; (2) Height detection mostly uses a single sensor, which is easily affected by crop residues, field vibration and dust, resulting in large detection errors and inability to accurately obtain the real-time height of the header off the ground; (3) Conventional PID control is a fixed parameter control, which is difficult to adapt to the nonlinear and time-varying characteristics caused by field terrain undulations and crop growth status differences, and has problems such as large overshoot, obvious steady-state error and poor adaptive ability. Summary of the Invention

[0004] The technical problem to be solved by this invention is to overcome the existing defects and provide an intelligent control method for the header height of a combine harvester. By improving the height detection accuracy through multi-sensor data fusion and optimizing the control response characteristics by combining a neural network fuzzy PID algorithm, the invention achieves real-time and precise control of the header height, thereby reducing harvest losses and improving operating efficiency, and can effectively solve the problems in the background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent control of the header height of a combine harvester, comprising the following steps:

[0006] S1: Establish the geometric model of the cutting platform motion and determine the functional relationship between the height of the cutting platform off the ground and the adjustment amount of the cutting platform hydraulic cylinder;

[0007] S2: A multi-sensor data fusion method based on BP neural network is used to preprocess and perform neural network operations on the data collected by the angle sensor to obtain the real-time ground clearance of the cutting platform;

[0008] S3: Construct a hierarchical control framework for the cutter height based on a BP neural network fuzzy PID algorithm. Perform fuzzy processing on the height deviation and the rate of change of deviation, adaptively adjust the PID control parameters, and realize real-time closed-loop control of the cutter height.

[0009] As a preferred embodiment of the present invention, the geometric model of the cutting table motion in step S1 satisfies the following geometric relationship:

[0010]

[0011]

[0012]

[0013]

[0014] Where ΔH represents the difference between the actual ground clearance of the cutting platform during operation and the initially set ground clearance, H is the actual cutter ground clearance, H1 is the height from the center of rotation of the cutting platform to the ground, H0 is the initial cutter ground clearance, α is the angle formed between the line connecting the mounting point of the cutting platform cylinder on the cutting platform and the center of rotation of the cutting platform, and the line connecting the cutter position and the center of rotation of the cutting platform, β is the angle formed between the mounting points of the two ends of the cutting platform cylinder and the line connecting the center of rotation of the cutting platform at the actual ground clearance, and β0 is the corresponding angle in the initial state, L1 represents the distance between the bridge contact point and the cutter position, L2 represents the distance from the bridge contact point to the mounting point of the cutting platform cylinder on the cutting platform, L3 represents the distance from the bridge contact point to the mounting point of the cutting platform cylinder on the chassis, L4 is the length between the two mounting points of the cutting platform cylinder at the actual ground clearance, L 40 Where ΔL is the initial cylinder length, and ΔL4 is the change in the cutting table cylinder length;

[0015] As a preferred embodiment of the present invention, step S2 specifically includes:

[0016] (1) Data preprocessing: The raw angle data collected by the four angle sensors are normalized to remove noise and interference. The processing formula is as follows:

[0017]

[0018] Among them, X min and X max These are the minimum and maximum values ​​in the sample data;

[0019] (2) Constructing a BP neural network: The input layer has four neurons, corresponding to the preprocessed angle data θ1, θ2, θ3, and θ4;

[0020] The hidden layer has four neurons, and the nonlinear transformation is completed by weighted summation, bias superposition and sigmoid activation function;

[0021] The output layer has one neuron, which uses a linear activation function to output the sprocket height control signal.

[0022] (3) Neural network training: using mean squared error as the loss function, the backpropagation algorithm combined with the stochastic gradient descent optimizer is used to iteratively update the network weights and biases so that the output approximates the actual target value of the cutter height.

[0023] As a preferred embodiment of the present invention, step S3 specifically includes:

[0024] (1) Construct a dual-input-three-output fuzzy controller and establish a fuzzy control model for the height of the cutter based on the Simulink fuzzy toolbox;

[0025] (2) Define the membership function: The basic universe of discourse for height deviation e and height deviation change rate ec is {−1,1}, and the Gaussian membership function is adopted;

[0026] The basic universe of discourse for the PID parameter corrections ΔKp, ΔKi, and ΔKd is {−1,1}, and trigonometric membership functions are used.

[0027] The fuzzy subsets are all {negative large (NB), negative small (NS), zero (Z), positive small (PS), positive large (PB)};

[0028] (3) Establish fuzzy control rules: adopt the "IF...AND...THEN..." rule to match the height adjustment amount of the cutting table according to the height deviation and the rate of change of deviation;

[0029] (4) Online control: By using fuzzy reasoning and defuzzification to output precise control quantities, the PID parameters Kp, Ki, and Kd are adaptively adjusted to achieve automatic closed-loop control of the stubble height of the header.

[0030] As a preferred embodiment of the present invention, the core of the fuzzy control rule includes:

[0031] If the height deviation e is negative and the height deviation change rate ec is positive, then the PID parameter correction is positive.

[0032] If the height deviation e is negatively small and the height deviation change rate ec is positively small, then the PID parameter correction amount is positively small.

[0033] If the height deviation e is zero and the height deviation change rate ec is zero, then the PID parameter correction is zero.

[0034] If the height deviation e is positively small and the height deviation change rate ec is negatively small, then the PID parameter correction amount is negatively small.

[0035] If the height deviation e is positive and the height deviation change rate ec is negative, then the PID parameter correction amount is negative.

[0036] Compared with the prior art, the beneficial effects of the present invention are: (1) High detection accuracy: The four-angle sensor collaborative perception + BP neural network data fusion effectively filters vibration and stubble interference, and the height detection error is significantly reduced; (2) Strong control adaptability: Based on the BP neural network fuzzy PID algorithm, the PID parameters are adjusted in real time to adapt to the nonlinear characteristics of terrain undulation and crop status, with small overshoot and low steady-state error; (3) Good operation effect: The cutting platform has high contour accuracy and good stubble height consistency, which greatly reduces the cutting platform loss of crops such as soybeans, improves the operation efficiency and working condition adaptability of the combine harvester, and is applicable to various field terrains such as flat and undulating, taking into account the height control needs of different crops. Attached Figure Description

[0037] Figure 1 This is a flowchart of the present invention;

[0038] Figure 2 This is a schematic diagram of the geometric model of the cutting table motion of the present invention;

[0039] Figure 3 This is a schematic diagram of the BP neural network structure of the present invention;

[0040] Figure 4 This is a block diagram of the control system structure of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] like Figure 1-4 This invention provides a technical solution: a method for intelligent control of the header height of a combine harvester, comprising the following steps:

[0043] S1: Establish the dynamic model of the header, that is, the geometric model of the header motion, and determine the functional mapping relationship between the header's ground clearance and the adjustment amount of the header cylinder;

[0044] S2: Design a BP neural network-based method for detecting the height of the cutting platform, which integrates multi-angle sensor data and obtains the real-time height of the cutting platform above the ground through preprocessing and neural network operation;

[0045] S3: Construct a hierarchical control framework for the cutter height based on a BP neural network fuzzy PID algorithm. Perform fuzzy processing on the height deviation and the rate of change of deviation, adaptively adjust the PID control parameters, and realize real-time closed-loop control of the cutter height.

[0046] In step S1, the geometric model of the cutting table motion satisfies the following relationship:

[0047]

[0048]

[0049] Among them, the change in the hydraulic cylinder of the cutting table:

[0050] Finally, the functional relationship between the change in the hydraulic cylinder of the header and the change in the height of the header off the ground is:

[0051]

[0052] In the formula: ΔH is the difference between the actual height of the cutting table from the ground during operation and the initially set height of the cutting table from the ground;

[0053] H represents the actual height of the cutter above the ground;

[0054] H1 is the height of the cutting table rotation center from the ground;

[0055] H0 is the initial cutter height above the ground;

[0056] α is the angle formed between the line connecting the mounting point of the hydraulic cylinder on the cutting table and the center of rotation of the cutting table, and the line connecting the position of the cutter and the center of rotation of the cutting table;

[0057] β represents the angle formed by the lines connecting the mounting points at both ends of the hydraulic cylinder of the header to the center of rotation of the header, when the header is at its actual height above the ground.

[0058] β0 is the corresponding included angle in the initial state;

[0059] L1 represents the distance between the bridge contact point and the cutter position;

[0060] L2 refers to the distance from the bridge contact point to the mounting point of the hydraulic cylinder on the cutting table;

[0061] L3 represents the distance from the bridge contact point to the mounting point of the cutting table cylinder on the chassis;

[0062] L4 is the length between the two mounting points of the hydraulic cylinder at the actual height of the cutting table above the ground;

[0063] L 40 This is the initial cylinder length;

[0064] ΔL4 represents the change in the hydraulic cylinder of the cutting table.

[0065] Step S2: Cutter height detection based on BP neural network, including:

[0066] (1) Data preprocessing: The raw angle data collected by the four angle sensors are normalized to remove noise interference and improve the processing effect of the neural network. The formula is:

[0067]

[0068] In the formula, X min and X max These are the minimum and maximum values ​​in the sample data;

[0069] (2) The BP neural network structure includes:

[0070] Input layer: Contains four neurons, corresponding to the preprocessed angle data θ1, θ2, θ3, and θ4, which together form the input vector. The role of the input layer is to introduce sensor data into the neural network, providing a foundation for subsequent processing;

[0071] Hidden layer: Contains four neurons. For each neuron in the hidden layer, its input is the weighted sum of the outputs of all neurons in the input layer plus a bias. Taking the i-th neuron in the hidden layer (i=1, 2, 3, 4) as an example, its input z i h The calculation formula is:

[0072]

[0073] The superscript "h" represents a hidden layer;

[0074] ω ij h It is the connection weight from the j-th neuron in the input layer to the i-th neuron in the hidden layer. It represents the degree of influence of the input data on that hidden layer neuron, for example, ω. 23 h This represents the connection weight from the 3rd neuron in the input layer to the 2nd neuron in the hidden layer;

[0075] b i h It is the bias of the i-th neuron in the hidden layer, used to adjust the activation threshold of the neuron.

[0076] Then, the hidden layer neurons will process the input z i h By applying a nonlinear transformation using an activation function, the output a of the neuron can be obtained. i h Commonly used activation functions include the Sigmoid function and the ReLU function. Here, we will take the Sigmoid function as an example. Its expression is:

[0077]

[0078] get:

[0079]

[0080] Through nonlinear transformations of activation functions, neural networks can learn complex patterns and features in input data;

[0081] Output layer: Contains one neuron, whose input is the weighted sum of the outputs of all neurons in the hidden layer plus a bias. The input z of the output layer neuron... o The calculation formula is:

[0082]

[0083] The superscript "o" represents the output layer;

[0084] w i o It represents the connection weight from the i-th neuron in the hidden layer to the neuron in the output layer;

[0085] b o It is the bias of the output layer neurons;

[0086] The output layer uses a linear activation function:

[0087] Therefore, the output y of the output layer is:

[0088]

[0089] In the formula, the output y is the control signal used to adjust the height of the cutting table.

[0090] (3) Neural network training

[0091] Neural network Xunliao adjusts connection weights and bias Adjust the network output y to be as close as possible to the actual header height target value. .

[0092] Mean squared error is used as the loss function:

[0093]

[0094] In the formula, N represents the number of training samples;

[0095] y k It is the neural network output of the k-th sample;

[0096] It is the actual target value for adjusting the header height corresponding to the kth sample;

[0097] It is the square of the error between the network output of the k-th sample and the target value.

[0098] The average error over the entire training set is obtained by summing the squared errors of all samples and taking the average.

[0099] During training, a commonly used training algorithm is backpropagation combined with a stochastic gradient descent optimizer. Specifically, in each iteration, the weights and biases are updated based on the gradient of the loss function with respect to the weights and biases. The update formula for the weights ω is:

[0100]

[0101] In the formula, ω represents the connection weights in the neural network, which can be the weights from the input layer to the hidden layer. ij h It can also be the weight ω from the hidden layer to the output layer. i o ;

[0102] ← indicates an update operation, which assigns the value calculated on the right side of the equals sign to the variable on the left side;

[0103] η is the learning rate, which is a hyperparameter used to control the step size of each parameter update;

[0104] It is the partial derivative of the loss function L(W,b) with respect to the weight ω, also known as the gradient.

[0105] By collecting data from four sensors and processing it through a neural network with a specific structure, the height of the cutting platform can be accurately calculated.

[0106] Step S3: High-level control based on BP neural network fuzzy PID, specifically including:

[0107] (1) Design a fuzzy controller. Use the graphical user interface in the Simulink fuzzy toolbox to create a fuzzy control model for the height of the cutter table. Use a dual-input-three-output controller.

[0108] (2) Membership function setting: The real-time height x of the cutter head is obtained by the above-mentioned BP neural network multi-sensor fusion processing to obtain a relatively accurate value. By comparing it with the cutter head height value of the set harvested crop, the deviation e and the deviation change rate ec are obtained. Based on the empirical value of the cutter head height suitable for different crops during the harvesting process of the harvester and by consulting relevant materials, the basic domain of the input and output variables is determined.

[0109] The universes of discourse for e and ec are {-1, 1}, and the universe of discourse for the output variable u is {-1, 1}.

[0110] The fuzzy control theory is used to correct it. The fuzzy subsets of the deviation e, the deviation rate ec, and the outputs ΔKp, ΔKi, and ΔKd are all {negative large (NB), negative small (NS), zero (Z), positive small (PS), positive large (PB)}. The input height deviation e and the input height deviation change rate ec adopt Gaussian functions, and the membership functions of ΔKp, ΔKi, and ΔKd all adopt trigonometric functions.

[0111] (3) Establish a fuzzy control rule table: Use fuzzy rules in the form of “IF…AND…THEN…” to establish the control strategy and summarize them into a fuzzy rule table for adjusting the three parameters of the PID controller.

[0112] Here are five fuzzy control rules:

[0113] If the height deviation e is negative and the height deviation change rate ec is positive, then the header height adjustment amount is positive. That is, when the actual header height is much lower than the appropriate height and the crop height is increasing rapidly, the header needs to be raised significantly to adapt to the crop height and ensure the harvesting effect.

[0114] If the height deviation e is negative and the height deviation change rate ec is positive, then the header height adjustment amount is positive. That is, when the actual header height is slightly lower than the appropriate height and the crop height increases slightly, the header needs to be raised slightly to reach the appropriate height.

[0115] If the height deviation e is zero and the height deviation change rate ec is zero, then the header height adjustment is zero. That is, when the actual header height is just at the appropriate height and the crop height has not changed, there is no need to adjust the header height; the current state can be maintained.

[0116] If the height deviation e is positively small and the height deviation change rate ec is negatively small, then the header height adjustment amount is negatively small. That is, when the actual header height is slightly higher than the appropriate height and the crop height decreases slightly, the header height needs to be reduced slightly to bring it back to the appropriate height.

[0117] If the height deviation e is positive and the height deviation change rate ec is negative, then the header height adjustment amount is negative. That is, when the actual header height is much higher than the appropriate height and the crop height is decreasing rapidly, the header height needs to be reduced significantly to adapt to the crop height and avoid problems such as missed cutting.

[0118] (4) Establishment of fuzzy PID control model based on BP neural network: According to the different deviation e and deviation change rate ec, the parameters Kp, Ki and Kd in the PID controller are adjusted online using fuzzy control algorithm, so as to realize the automatic control of the stubble height of the cutting platform.

[0119] In fuzzy engineering, precise input values ​​are quantized and converted into fuzzy input values. Then, fuzzy control calculation rules are used to perform fuzzy inference. After fuzzy analysis, the fuzzy values ​​are converted into precise values ​​and sent to the controlled object.

[0120] To address the varying height adjustment requirements of the header in different operating conditions of combine harvesters, the above method employs a detection scheme that integrates collaborative sensing from four-angle sensors (cutter blade, side contour plates, and header body) with BP neural network data fusion. Based on sensor characteristics and operating condition adaptation rules, multi-source detection data is processed in layers. A preprocessing model handles noise filtering and data normalization, while the BP neural network handles operating condition identification and real-time height calculation. Furthermore, a neural network fuzzy PID algorithm adaptively tunes control parameters while constraining the maximum extension and retraction rate of the header cylinder. This minimizes control errors caused by environmental interference and nonlinear characteristics, thereby improving header contouring accuracy and stubble height consistency, significantly reducing header losses during soybean harvesting, and fully enhancing the combine harvester's operating efficiency and adaptability.

[0121] The parts of the invention not described in detail are prior art. Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent control of header height in a combine harvester, characterized in that: Includes the following steps: S1: Establish the geometric model of the cutting platform motion and determine the functional relationship between the height of the cutting platform off the ground and the adjustment amount of the cutting platform hydraulic cylinder; S2: A multi-sensor data fusion method based on BP neural network is used to preprocess and perform neural network operations on the data collected by the angle sensor to obtain the real-time ground clearance of the cutting platform; S3: Construct a hierarchical control framework for the cutter height based on a BP neural network fuzzy PID algorithm. Perform fuzzy processing on the height deviation and the rate of change of deviation, adaptively adjust the PID control parameters, and realize real-time closed-loop control of the cutter height.

2. The intelligent control method for header height of a combine harvester according to claim 1, characterized in that: The geometric model of the cutting table motion in step S1 satisfies the following geometric relationship: Where ΔH represents the difference between the actual ground clearance of the cutting platform during operation and the initially set ground clearance, H is the actual cutter ground clearance, H1 is the height from the center of rotation of the cutting platform to the ground, H0 is the initial cutter ground clearance, α is the angle formed between the line connecting the mounting point of the cutting platform cylinder on the cutting platform and the center of rotation of the cutting platform, and the line connecting the cutter position and the center of rotation of the cutting platform, β is the angle formed between the mounting points of the two ends of the cutting platform cylinder and the line connecting the center of rotation of the cutting platform at the actual ground clearance, and β0 is the corresponding angle in the initial state, L1 represents the distance between the bridge contact point and the cutter position, L2 represents the distance from the bridge contact point to the mounting point of the cutting platform cylinder on the cutting platform, L3 represents the distance from the bridge contact point to the mounting point of the cutting platform cylinder on the chassis, L4 is the length between the two mounting points of the cutting platform cylinder at the actual ground clearance, L 40 ΔL4 represents the initial cylinder length, and ΔL4 represents the change in the hydraulic cylinder length of the cutting table.

3. The intelligent control method for header height of a combine harvester according to claim 1, characterized in that: Step S2 specifically includes: (1) Data preprocessing: The raw angle data collected by the four angle sensors are normalized to remove noise and interference. The processing formula is as follows: Among them, X min and X max These are the minimum and maximum values ​​in the sample data; (2) Constructing a BP neural network: The input layer has four neurons, corresponding to the preprocessed angle data θ1, θ2, θ3, and θ4; The hidden layer has four neurons, and the nonlinear transformation is completed by weighted summation, bias superposition and sigmoid activation function; The output layer has one neuron, which uses a linear activation function to output the sprocket height control signal. (3) Neural network training: using mean squared error as the loss function, the backpropagation algorithm combined with the stochastic gradient descent optimizer is used to iteratively update the network weights and biases so that the output approximates the actual target value of the cutter height.

4. The intelligent control method for header height of a combine harvester according to claim 1, characterized in that: Step S3 specifically includes: (1) Construct a dual-input-three-output fuzzy controller and establish a fuzzy control model for the height of the cutter based on the Simulink fuzzy toolbox; (2) Define the membership function: The basic universe of discourse for height deviation e and height deviation change rate ec is {−1,1}, and the Gaussian membership function is adopted; The basic universe of discourse for the PID parameter corrections ΔKp, ΔKi, and ΔKd is {−1,1}, and trigonometric membership functions are used. The fuzzy subsets are all {negative large (NB), negative small (NS), zero (Z), positive small (PS), positive large (PB)}; (3) Establish fuzzy control rules: adopt the "IF...AND...THEN..." rule, and match the height adjustment amount of the cutting table according to the height deviation and the rate of change of deviation; (4) Online control: By using fuzzy reasoning and defuzzification to output precise control quantities, the PID parameters Kp, Ki, and Kd are adaptively adjusted to achieve automatic closed-loop control of the stubble height of the header.

5. The intelligent control method for header height of a combine harvester according to claim 4, characterized in that: The core of the fuzzy control rules includes: If the height deviation e is negative and the height deviation change rate ec is positive, then the PID parameter correction is positive. If the height deviation e is negatively small and the height deviation change rate ec is positively small, then the PID parameter correction amount is positively small. If the height deviation e is zero and the height deviation change rate ec is zero, then the PID parameter correction is zero. If the height deviation e is positively small and the height deviation change rate ec is negatively small, then the PID parameter correction amount is negatively small. If the height deviation e is positive and the height deviation change rate ec is negative, then the PID parameter correction amount is negative.

Images