A method for predicting and regulating the working walking speed of a rice-wheat combine harvester

Abstract

A method for predicting and controlling the walking speed of a rice-wheat combine harvester is proposed. Based on the harvester's walking speed, a bootstrap algorithm is used to sample speed expansion data. A fully connected generative adversarial network (GAN) model is constructed, and a multivariate sequence data generation model is obtained by training the GAN model. The expanded speed samples are input into the multivariate sequence data generation model to generate corresponding expanded multimodal samples. A random forest algorithm is used to construct a walking speed prediction model for the rice-wheat combine harvester. The expanded multimodal samples are input into the walking speed prediction model for random forest training to obtain the optimal walking speed prediction model. Real-time data from the rice-wheat combine harvester is input into the optimal walking speed prediction model to obtain the optimal speed prediction value. Finally, incremental PID control is used to regulate the walking speed of the rice-wheat combine harvester. This method has the advantages of high speed rationality, high algorithm robustness, and good real-time speed prediction.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology for rice and wheat combine harvesters, and in particular to a method for predicting and controlling the operating speed of a rice and wheat combine harvester. Background Technology

[0002] Automatic navigation for rice and wheat combine harvesters has developed rapidly as a key technology for the construction of unmanned and smart farms. However, the actual navigation speed of the harvester is still a fixed empirical value or a low value, resulting in poor adaptability to working conditions, low operating efficiency, and the risk of machine blockage.

[0003] In existing technologies, research on the prediction and control of combine harvester operating speed mainly relies on the estimation of the harvester's feed rate, which can be divided into two categories: 1. Feed rate estimation based on crop attributes (i.e., the state of the crop in front of the harvester); 2. Feed rate estimation based on mechanical conditions (i.e., harvester load, speed and torque of key rotating parts of the harvester). However, when estimating feed rate based on crop conditions, it is difficult to establish a mapping relationship between crop image information and feed rate, resulting in large errors in feed rate estimation; when estimating feed rate based on the speed and torque of key rotating parts, there is a problem of feed rate estimation delay. The above methods and existing problems will all have an adverse impact on the prediction of the optimal operating speed of the combine harvester, resulting in inaccurate speed prediction. Mismatch in operating speed can lead to wasted harvester power or blockage, leaving potential risks for the harvester's automatic navigation operation. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting and controlling the operating speed of a rice and wheat combine harvester, which overcomes the limitations of traditional harvester speed prediction based on feed rate estimation, and has high speed rationality, high algorithm robustness, and good real-time speed prediction.

[0005] The technical solution adopted by this invention to solve the above-mentioned technical problems is: a method for predicting and controlling the operating speed of a rice and wheat combine harvester, comprising the following steps:

[0006] Step 1: The rice and wheat combine harvester operates under various working conditions. The walking speed of the rice and wheat combine harvester is recorded. The actual harvesting width, rice and wheat ear density, actual cutting height, straw moisture content, threshing drum speed, header spindle speed, and bridge conveyor chain rake spindle speed are obtained through the detection data of multiple sensors.

[0007] Step 2: Obtain the minimum value P and maximum value N of the original speed based on the walking speed of the rice and wheat combine harvester. Then, set the speed expansion area, with the lower limit L N. Then, use the self-service Bootstrap algorithm to sample the speed expansion data.

[0008] New velocity samples are generated using a uniform distribution within the data diffusion intervals [L, P] and [N, U].

[0009] In the original velocity range [P, N], the original sample distribution is first fitted, and the frequency distribution of each velocity range is counted at fixed intervals, denoted as C = {C1, C2, C3, ..., C...}. n};

[0010] Then calculate the frequency count and the maximum frequency C. max The ratio of R to R is denoted as R = {R1, R2, R3, ..., R...}. n}, calculate the median value of each speed interval and C max The absolute difference of the midpoints of the corresponding intervals is denoted as D = {D1, D2, D3, ..., D...} n};

[0011] Then, random samples are generated for each speed range. The formula for calculating the sample generation quantity N is:

[0012] (1);

[0013] Finally, in the original data interval [P, N], new velocity samples are generated again using a uniform distribution.

[0014] The set of all newly generated velocity samples in all intervals is used as the expanded velocity sample;

[0015] Step 3: Construct a Generative Adversarial Network (GAN) model using a fully connected layer. The GAN model includes a generator and a discriminator. The generator includes a batch normalization layer and a linear rectified activation function, and introduces nonlinear factors. The last layer of the generator applies a hyperbolic tangent activation function for data scaling.

[0016] The input dimension of the discriminator is matched with the output dimension of the generator and both are set to 7, corresponding to a multimodal sample consisting of actual harvest width, rice and wheat ear density, actual cutting height, straw moisture content, threshing drum speed, header spindle speed, and bridge conveyor chain rake spindle speed.

[0017] The discriminator's output layer uses the sigmoid activation function to generate probability values ​​between 0 and 1, which represent the likelihood that the input data is real data.

[0018] The GAN model is trained by taking random noise data as the initial input to the generator and outputting multivariate data through the generator. The binary cross-entropy loss is selected as the loss function. The loss is calculated based on the probability value output by the discriminator and the true label of the data sample. The multivariate sequence data generation model is obtained through iterative optimization.

[0019] Step 4: Input the expanded speed samples obtained in Step 2 into the multivariate sequence data generation model to generate expanded multimodal samples corresponding to the expanded speed samples, including 7 types of amplified sample data: actual harvesting width, rice and wheat ear density, actual cutting height, straw moisture content, threshing drum speed, header spindle speed, and bridge conveyor chain harrow spindle speed.

[0020] Step 5: Construct a walking speed prediction model for rice and wheat combine harvesters using the random forest algorithm. Input extended multimodal samples into the walking speed prediction model for random forest training. Use the Bayesian algorithm to optimize the number of decision trees, the maximum depth of decision trees, and the maximum number of leaf nodes to obtain the optimal walking speed prediction model.

[0021] Step 6: Smooth the real-time travel speed of the rice and wheat combine harvester:

[0022] v r,t =D[0]×v r,t0 +D[1]×v r,t-1 +D[2]×v r,t-2 +D[3]×v r,t-3 +D[4]×v r,t-4 (2);

[0023] In equation (2), D = [0.5, 0.2, 0.1, 0.1, 0.1] represents the velocity smoothing parameter, v r,t The smoothed velocity, v r,t0 Let v be the velocity at the current moment. r,t-1 The velocity at the previous moment;

[0024] Step 7: Input the real-time data of the rice and wheat combine harvester into the optimal walking speed prediction model to obtain the optimal speed prediction value, and then use incremental PID to regulate the walking speed of the rice and wheat combine harvester.

[0025] The output of the incremental PID controller is the control voltage increment. The input is the difference between the smoothed actual speed and the optimal speed prediction. The formula for the incremental PID controller is:

[0026] (3);

[0027] In equation (3), The error at the current moment, This represents the error from the previous moment; This is the error from the first two time points. To control the increment of the quantity, K p K i K dThese are proportional gain, integral gain, and derivative gain, respectively.

[0028] The incremental PID controller, combined with the control voltage from the previous moment... Get the control voltage at the current moment The vehicle controller receives and The corresponding optimal speed is used to control the operating speed of the rice and wheat combine harvester.

[0029] Preferably, in step 1, the walking speed is recorded using an RTK-GNSS device; the multiple sensors include a straw moisture content detection sensor, a threshing drum speed sensor, a bridge conveyor chain rake main shaft speed sensor, a header main shaft speed sensor, and a header height sensor.

[0030] Preferably, the straw moisture content sensor is installed on the inner wall of the bridge conveyor, and the threshing drum speed sensor, the bridge conveyor chain rake main shaft speed sensor, the header main shaft speed sensor, and the header height sensor are respectively installed at the corresponding detection positions of the rice and wheat combine harvester.

[0031] Preferably, in step 7, the output time interval of the optimal speed is 2s, the speed is the average value of the system output speed within 2s, and the trigger interval of the incremental PID controller is 0.5s.

[0032] Preferably, after obtaining the optimal speed in step 7, the optimal speed v is... p,t Compared with the output speed v at the previous moment p,t-1 Perform smoothing, set a speed limit, and apply weighted control.

[0033] Preferably, the weighted control method involves collecting real-time threshing loss data from rice and wheat combine harvesters and classifying it into three categories: low, medium, and high. Then, the optimal operating travel speed is optimized for each category based on different weighting coefficients.

[0034] Low: Working walking speed = optimal working walking speed y × 1.1;

[0035] In the middle: the working walking speed = the optimal working walking speed y;

[0036] High: Working walking speed = optimal working walking speed y × 0.75.

[0037] According to the above technical solution, the beneficial effects of the present invention are:

[0038] This invention proposes an "end-to-end" optimal operating speed prediction model for harvesters based on multimodal information fusion. This model overcomes the limitations of traditional harvester speed prediction based on feed rate estimation and optimizes the optimal operating speed by combining harvesting quality. It has the advantages of high speed rationality, high algorithm robustness, and good real-time speed prediction. Attached Figure Description

[0039] Figure 1 This is a block diagram of the method of the present invention;

[0040] Figure 2 This is the histogram of the original velocity sample distribution. Detailed Implementation

[0041] This embodiment provides a method for predicting and controlling the operating speed of a rice and wheat combine harvester, such as... Figure 1 As shown, the specific steps include:

[0042] Step 1: The rice and wheat combine harvester operates under various working conditions. The traveling speed of the rice and wheat combine harvester is recorded by RTK-GNSS equipment, and the actual harvesting width, rice and wheat ear density, actual cutting height, straw moisture content, threshing drum speed, header spindle speed, and bridge conveyor chain rake spindle speed are obtained through the detection data of multiple sensors.

[0043] The multiple sensors used include: a straw moisture content detection sensor, installed on the inner wall of the bridge conveyor; a threshing drum speed sensor, a bridge conveyor chain rake main shaft speed sensor, a header main shaft speed sensor, and a header height sensor, which are installed at their respective detection positions on the rice and wheat combine harvester.

[0044] Step 2: Assume the original velocity dataset is Y = {y1, y2, …, y} n Distribution analysis of velocity data, such as Figure 2 As shown, the minimum value P and maximum value N of the original speed are obtained based on the walking speed of the rice and wheat combine harvester. Then, a speed expansion region is set, with the lower limit L N. Then, the speed expansion data is sampled using the self-service Bootstrap algorithm.

[0045] In the data diffusion intervals [L, P] and [N, U], new velocity samples are generated using a uniform distribution with a sample interval of 0.04 km / h.

[0046] Within the original speed range [P, N], we first fit the original sample distribution, and then statistically analyze the frequency distribution of each speed range at fixed intervals of 0.20 km / h, denoted as C = {C1, C2, C3, ..., C...}. n}

[0047] Then calculate the frequency count and the maximum frequency C. max The ratio of R to R is denoted as R = {R1, R2, R3, ..., R...}. n}, calculate the median value of each speed interval and C max The absolute difference of the midpoints of the corresponding intervals is denoted as D = {D1, D2, D3, ..., D...}n}

[0048] Then, random samples are generated for each speed range. The formula for calculating the sample generation quantity N is:

[0049] (1).

[0050] Finally, in the original data interval [P, N], new velocity samples are generated again using a uniform distribution, with a sample interval of 0.04 km / h.

[0051] The set of all newly generated velocity samples in all intervals is used as the expanded velocity sample.

[0052] Step 3: Construct a Generative Adversarial Network (GAN) model using a fully connected layer. The GAN model includes a generator and a discriminator. The generator includes a batch normalization layer and a linear rectified activation function, and introduces nonlinear factors. The last layer of the generator applies a hyperbolic tangent activation function for data scaling.

[0053] The input dimension of the discriminator is matched with the output dimension of the generator and both are set to 7, corresponding to the multimodal samples composed of actual harvesting width, rice and wheat ear density, actual cutting height, straw moisture content, threshing drum speed, header spindle speed, and bridge conveyor chain rake spindle speed.

[0054] The discriminator's output layer uses the sigmoid activation function to generate probability values ​​between 0 and 1, representing the likelihood that the input data is real data.

[0055] The GAN model is trained by using random noise data as the initial input to the generator, which outputs multivariate data. The binary cross-entropy loss is selected as the loss function. The loss is calculated based on the probability value output by the discriminator and the true label of the data sample. The multivariate sequence data generation model is obtained through iterative optimization.

[0056] Step 4: Input the expanded velocity samples obtained in Step 2 into the multivariate sequence data generation model to generate expanded multimodal samples corresponding to the expanded velocity samples, including 7 types of amplified sample data: actual harvesting width, rice and wheat ear density, actual cutting height, straw moisture content, threshing drum speed, header spindle speed, and bridge conveyor chain harrow spindle speed.

[0057] Step 5: Construct a walking speed prediction model for rice and wheat combine harvesters using the random forest algorithm. Input extended multimodal samples into the walking speed prediction model for random forest training. Use the Bayesian algorithm to optimize the number of decision trees, the maximum depth of decision trees, and the maximum number of leaf nodes to obtain the optimal walking speed prediction model.

[0058] Step 6: Smooth the real-time travel speed of the rice and wheat combine harvester:

[0059] v r,t =D[0]×v r,t0 +D[1]×v r,t-1 +D[2]×v r,t-2 +D[3]×v r,t-3 +D[4]×v r,t-4 (2);

[0060] In equation (2), D = [0.5, 0.2, 0.1, 0.1, 0.1] represents the velocity smoothing parameter, v r,t The smoothed velocity, v r,t0 Let v be the velocity at the current moment. r,t-1 The velocity at the previous moment.

[0061] Step 7: Input the real-time data of the rice and wheat combine harvester into the optimal walking speed prediction model to obtain the optimal speed prediction value, and then use incremental PID to regulate the walking speed of the rice and wheat combine harvester.

[0062] The output of the incremental PID controller is the control voltage increment. The input is the difference between the smoothed actual speed and the optimal speed prediction. The formula for the incremental PID controller is:

[0063] (3);

[0064] In equation (3), The error at the current moment, This represents the error from the previous moment; This is the error from the first two time points. To control the increment of the quantity, K p K i K d These are proportional gain, integral gain, and derivative gain, respectively.

[0065] The incremental PID controller, combined with the control voltage from the previous moment... Get the control voltage at the current moment The vehicle controller receives and The corresponding optimal speed is used to control the operating speed of the rice and wheat combine harvester.

[0066] Because harvesters are large in mass and have high inertia, and hydraulic chassis control has a certain lag, large and frequent speed adjustments during harvesting operations can cause the harvester to "jerk," affecting operational stability and even leading to mechanical failure. Therefore, after obtaining the optimal speed in step 7, the optimal speed v is... p,tCompared with the output speed v at the previous moment p,t-1 Perform smoothing and set a speed limit. The smoothing rules are as follows:

[0067] if vp,t> 5.5;

[0068] vp,t = 5.5;

[0069] if |vp,t - vp,t-1| > 0.4;

[0070] vp,t = (vp,t + vp,t-1) / 2;

[0071] Considering the steady-state time of the incremental PID model and the hysteresis of the hydraulic chassis control, the output time interval of the optimal speed prediction model is set to 2s, the speed is the average value of the system output speed within 2s, and the trigger interval of the incremental PID is 0.5s.

[0072] Simultaneously, weighted control is applied to the optimal speed. The method involves real-time collection of threshing loss data from the rice and wheat combine harvester, categorizing it into low, medium, and high levels. Then, the optimal operating travel speed is optimized for each category based on different weighting coefficients.

[0073] Low: Working walking speed = optimal working walking speed y × 1.1;

[0074] In the middle: the working walking speed = the optimal working walking speed y;

[0075] High: Working walking speed = optimal working walking speed y × 0.75.

[0076] This embodiment proposes an "end-to-end" harvester optimal operating speed prediction model based on multimodal information fusion, which gets rid of the limitations of traditional harvester speed prediction based on feed rate estimation, and optimizes the optimal operating speed by combining harvesting operation quality. It has the advantages of high speed rationality, high algorithm robustness, and good real-time speed prediction.

Claims

1. A method for predicting and controlling the operating speed of a rice and wheat combine harvester, characterized in that, Includes the following steps: Step 1: The rice and wheat combine harvester operates under various working conditions. The walking speed of the rice and wheat combine harvester is recorded. The actual harvesting width, rice and wheat ear density, actual cutting height, straw moisture content, threshing drum speed, header spindle speed, and bridge conveyor chain rake spindle speed are obtained through the detection data of multiple sensors. Step 2: Obtain the minimum value P and maximum value N of the original speed based on the walking speed of the rice and wheat combine harvester. Then, set the speed expansion area, with the lower limit L N. Then, use the self-service Bootstrap algorithm to sample the speed expansion data. New velocity samples are generated using a uniform distribution within the data diffusion intervals [L, P] and [N, U]. In the original velocity range [P, N], the original sample distribution is first fitted, and the frequency distribution of each velocity range is counted at fixed intervals, denoted as C = {C1, C2, C3, ..., C...}. n }; Then calculate the frequency count and the maximum frequency C. max The ratio of R to R is denoted as R = {R1, R2, R3, ..., R...}. n }, calculate the median value of each speed interval and C max The absolute difference of the midpoints of the corresponding intervals is denoted as D = {D1, D2, D3, ..., D...} n }; Then, random samples are generated for each speed range. The formula for calculating the sample generation quantity N is: （1）； Finally, in the original data interval [P, N], new velocity samples are generated again using a uniform distribution. The set of all newly generated velocity samples in all intervals is used as the expanded velocity sample; Step 3: Construct a Generative Adversarial Network (GAN) model using a fully connected layer. The GAN model includes a generator and a discriminator. The generator includes a batch normalization layer and a linear rectified activation function, and introduces nonlinear factors. The last layer of the generator applies a hyperbolic tangent activation function for data scaling. The input dimension of the discriminator is matched with the output dimension of the generator and both are set to 7, corresponding to a multimodal sample consisting of actual harvest width, rice and wheat ear density, actual cutting height, straw moisture content, threshing drum speed, header spindle speed, and bridge conveyor chain rake spindle speed. The discriminator's output layer uses the sigmoid activation function to generate probability values ​​between 0 and 1, which represent the likelihood that the input data is real data. The GAN model is trained by taking random noise data as the initial input to the generator and outputting multivariate data through the generator. The binary cross-entropy loss is selected as the loss function. The loss is calculated based on the probability value output by the discriminator and the true label of the data sample. The multivariate sequence data generation model is obtained through iterative optimization. Step 4: Input the expanded speed samples obtained in Step 2 into the multivariate sequence data generation model to generate expanded multimodal samples corresponding to the expanded speed samples, including 7 types of amplified sample data: actual harvesting width, rice and wheat ear density, actual cutting height, straw moisture content, threshing drum speed, header spindle speed, and bridge conveyor chain harrow spindle speed. Step 5: Construct a walking speed prediction model for rice and wheat combine harvesters using the random forest algorithm. Input extended multimodal samples into the walking speed prediction model for random forest training. Use the Bayesian algorithm to optimize the number of decision trees, the maximum depth of decision trees, and the maximum number of leaf nodes to obtain the optimal walking speed prediction model. Step 6: Smooth the real-time travel speed of the rice and wheat combine harvester: v r,t =D[0]×v r,t0 +D[1]×v r,t-1 +D[2]×v r,t-2 +D[3]×v r,t-3 +D[4]×v r,t-4 (2); In equation (2), D = [0.5, 0.2, 0.1, 0.1, 0.1] represents the velocity smoothing parameter, v r,t The smoothed velocity, v r,t0 Let v be the velocity at the current moment. r,t-1 The velocity at the previous moment; Step 7: Input the real-time data of the rice and wheat combine harvester into the optimal walking speed prediction model to obtain the optimal speed prediction value, and then use incremental PID to regulate the walking speed of the rice and wheat combine harvester. The output of the incremental PID controller is the control voltage increment. The input is the difference between the smoothed actual speed and the optimal speed prediction. The formula for the incremental PID controller is: （3）； In equation (3), The error at the current moment, This represents the error from the previous moment; This is the error from the first two time points. To control the increment of the quantity, K p K i K d These are proportional gain, integral gain, and derivative gain, respectively. The incremental PID controller, combined with the control voltage from the previous moment... Get the control voltage at the current moment The vehicle controller receives and The corresponding optimal speed is used to control the operating speed of the rice and wheat combine harvester.

2. The method for predicting and controlling the operating speed of a rice and wheat combine harvester according to claim 1, characterized in that: In step 1, the walking speed is recorded using an RTK-GNSS device; the multiple sensors include a straw moisture content detection sensor, a threshing drum speed sensor, a bridge conveyor chain rake main shaft speed sensor, a header main shaft speed sensor, and a header height sensor.

3. The method for predicting and controlling the operating speed of a rice and wheat combine harvester according to claim 2, characterized in that: The straw moisture content sensor is installed on the inner wall of the bridge conveyor, while the threshing drum speed sensor, the bridge conveyor chain rake main shaft speed sensor, the header main shaft speed sensor, and the header height sensor are installed at their respective detection positions on the rice and wheat combine harvester.

4. The method for predicting and controlling the operating speed of a rice and wheat combine harvester according to claim 1, characterized in that: In step 7, the optimal speed output time interval is 2s, the speed is the average value of the system output speed within 2s, and the trigger interval of the incremental PID controller is 0.5s.

5. The method for predicting and controlling the operating speed of a rice and wheat combine harvester according to claim 1, characterized in that: After obtaining the optimal speed in step 7, the optimal speed v is... p,t Compared with the output speed v at the previous moment p,t-1 Perform smoothing, set a speed limit, and apply weighted control.

6. The method for predicting and controlling the operating speed of a rice and wheat combine harvester according to claim 5, characterized in that: The weighted control method involves collecting real-time threshing loss data from rice and wheat combine harvesters and categorizing it into low, medium, and high levels. Then, the optimal operating travel speed is optimized for each category based on different weighting coefficients. Low: Working walking speed = optimal working walking speed y × 1.1; In the middle: the working walking speed = the optimal working walking speed y; High: Working walking speed = optimal working walking speed y × 0.75.

Images