Harvester operation speed instruction feedback system

Abstract

The invention discloses a harvester operation speed instruction feedback system, which relates to the technical field of harvesters and comprises a feed quantity monitoring module, a controller and a rotating speed adjusting module. The feeding quantity monitoring module adopts a multi-mode sensing fusion technology, and integrates a mechanical sensor group and an optical sensor group; through cooperative control of multi-mode sensing fusion and an intelligent algorithm, the performance of the harvester is remarkably improved, the feeding quantity monitoring precision is improved by 20%-30%, the grain loss rate is reduced by 5%-8%, and grain waste caused by blockage and incomplete threshing is reduced; a fault tolerance mechanism is built in the system, when a traditional sensor is abnormal, optical monitoring is automatically switched, and the maintenance cost is reduced by 30%-40%; the rotating speed coupling relation of the driving wheel, the reel and the conveying belt is dynamically optimized through reinforcement learning and a multi-agent model, the operation speed control precision reaches + / -0.5 km / h, the equipment stability is improved by 40%-50%, meanwhile, the service life of key components is prolonged, and agricultural mechanization is promoted to be efficient, intelligent and low in consumption.

Description

Technical Field

[0001] The present invention relates to the technical field of harvesters, and particularly relates to a harvester operation speed command feedback system. Background Art

[0002] A harvester is a core device for efficiently harvesting crops in modern agriculture. By integrating structures such as a cutting table, a conveyor belt, and a threshing device, it realizes the integrated operation of crop cutting, conveying, threshing, and storage, and is widely used in crops such as wheat, rice, and corn, greatly improving agricultural production efficiency and reducing the dependence on labor.

[0003] Existing harvesters mainly rely on a mechanical transmission system: the driving wheels drive the whole machine to move, the reel guides the crops to the cutting table for cutting, the cut crops are transported to the threshing device by the conveyor belt to complete threshing, the grains are stored and then output, and the straw is separated and discharged through a sieve; during the operation process, speed control and component coordination are mostly adjusted based on manual experience. For example, the transmission ratio is manually adjusted or the driving speed is temporarily changed to adapt to the change of crop density.

[0004] Existing harvesters have significant defects: firstly, the monitoring of the feeding volume depends on traditional mechanical sensors, which are easily affected by component wear and measurement errors, resulting in lagging data and large errors; secondly, there is a lack of intelligent control, and it is impossible to dynamically adjust the operation speed and component rotation speed in real time according to the crop density, leading to frequent problems such as blockage and incomplete threshing; thirdly, the system stability is poor, and a single sensor failure will cause the operation to interrupt, resulting in high maintenance costs; for example, when the conveyor belt pressure sensor fails, the system cannot automatically switch to the standby monitoring mode, causing the feeding volume to get out of control and seriously affecting the operation continuity. Summary of the Invention

[0005] The purpose of the present invention is to provide a harvester operation speed command feedback system to solve the technical problem in the prior art that the monitoring of the feeding volume depends on traditional mechanical sensors, which are easily affected by component wear and measurement errors, resulting in lagging data and large errors.

[0006] The technical problems to be solved by the present invention can be realized through the following technical solutions:

[0007] A harvesting machine operating speed command feedback system, comprising a feeding amount monitoring module, a controller, and a speed regulation module; the feeding amount monitoring module adopts multi-modal sensing fusion technology, integrates a mechanical mechanics sensor group and an optical sensor group, and realizes cross-modal data correlation analysis through a Transformer architecture; the controller is built-in with a reinforcement learning algorithm and a PID control strategy, receives the data of the feeding amount monitoring module, and dynamically generates the target operating speed and the target speeds of each component, wherein the reinforcement learning algorithm takes maximizing the harvesting efficiency and minimizing the grain loss as the optimization objectives, and the Transformer module processes the historical operating data and the real-time working condition data to optimize the control strategy; the speed regulation module works in coordination based on a multi-agent reinforcement learning model and a PID controller, and each agent in the multi-agent model corresponds to a reel, a conveyor belt, or a drum, and optimizes the communication between agents through a Transformer to realize the dynamic regulation of the rotational speed coupling relationship of multiple components.

[0008] As a further solution of the present invention: the working process of the data fusion unit of the feeding amount monitoring module includes: when the data error of the traditional mechanical sensor exceeds a preset threshold, it automatically switches to the optical sensor data, extracts the lidar point cloud features through a convolutional neural network, and fuses the time series torque data and the spatial point cloud data by using the self-attention mechanism of the Transformer, and outputs a feeding amount prediction value with an error ≤ 5%.

[0009] As a further solution of the present invention: the system is applicable to wheat, rice, corn, and soybean harvesting scenarios, reduces the grain loss rate by 5% - 8% by dynamically adjusting the operating speed, and reduces the maintenance cost by 30% - 40%.

[0010] A method for controlling the operating speed of a harvesting machine, comprising a harvesting machine body; a cutting table is hinged to the front end of the harvesting machine body; a reel is installed above the front of the cutting table; driving wheels are symmetrically installed on both sides of the bottom of the harvesting machine body; a conveyor belt is arranged behind the cutting table; a threshing device is fixedly connected to the middle of the harvesting machine body, and the threshing device is linked to the power output end of the engine through a conveyor belt, and its outlet is connected to a storage bin at the rear of the machine body; the feeding amount data of the traditional sensor and the optical sensor are collected in real time, the multi-source data are fused through a Transformer architecture and the future feeding amount trend is predicted, the target operating speed and the target speeds of each component are generated based on the reinforcement learning algorithm, the single-component closed-loop regulation is realized through a PID controller, and at the same time, the rotational speed coupling relationship of the driving wheels, the reel, and the conveyor belt is coordinated through a multi-agent model, wherein the communication logic between multi-agents is optimized through a Transformer, dynamically responds to the change of crop density, and ensures the matching of the driving speed of the driving wheels, the rotational speed of the reel, and the conveying speed of the conveyor belt, so as to avoid crop accumulation or incomplete threshing.

[0011] As a further solution of the present invention: The collaborative control of the multi-agent model includes: Each agent independently controls a transmission component. The driving wheel agent adjusts the conveyor belt speed according to the driving speed. The reel agent receives the global operation status information through the Transformer and optimizes the matching relationship between the rotation speed and the driving speed in real time to ensure a stable feeding volume.

[0012] As a further solution of the present invention: The controller is built-in with a fault tolerance mechanism. When any sensor data in the traditional sensor group is abnormal, it automatically switches to the monitoring mode dominated by the optical sensor, and compensates for the data deviation through the deep learning algorithm to maintain the continuous operation of the system.

[0013] As a further solution of the present invention: The PID controller parameters of the rotation speed adjustment module are dynamically optimized by reinforcement learning. Combining the historical operation data of the multi-agent model, the proportional, integral, and differential coefficients are adaptively adjusted to achieve a rotation speed control accuracy of ±5 r / min.

[0014] As a further solution of the present invention: The Transformer architecture adopts a multi-head self-attention mechanism, and the number of heads is set to 8. The input data includes sensor time series data, point cloud spatial features, and historical fault records, and the output is the fused feeding volume prediction value and control instructions.

[0015] The beneficial effects of the present invention: Through the collaborative control of multi-modal sensing fusion and intelligent algorithms, the performance of the harvester is significantly improved: the monitoring accuracy of the feeding volume is increased by 20%-30%, the grain loss rate is reduced by 5%-8%, and the grain waste caused by blockage and incomplete threshing is reduced; The system is built-in with a fault tolerance mechanism, which automatically switches to optical monitoring when the traditional sensor is abnormal, and the maintenance cost is reduced by 30%-40%; Reinforcement learning and the multi-agent model dynamically optimize the rotational speed coupling relationship of the driving wheel, reel, and conveyor belt, and the operating speed control accuracy reaches ±0.5 km / h, and the equipment stability is improved by 40%-50%. At the same time, the service life of key components is extended, promoting the development of agricultural mechanization towards high efficiency, intelligence, and low consumption. Brief Description of the Drawings

[0016] The present invention will be further described below with reference to the accompanying drawings.

[0017] Figure 1 is a schematic diagram of the overall structure of the present invention;

[0018] In the figure: 1. Harvester body; 2. Driving wheel; 3. Reel; 4. Cutter bar; 5. Conveyor belt; 6. Threshing device. Detailed Embodiments

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0020] As Figure 1 shown, a feedback system for the operating speed command of a harvester includes a feeding amount monitoring module, a controller, and a speed regulation module; the feeding amount monitoring module adopts multi-modal sensing fusion technology, integrates a mechanical mechanics sensor group and an optical sensor group, and realizes cross-modal data correlation analysis through a Transformer architecture; the controller is built-in with a reinforcement learning algorithm and a PID control strategy, receives the data of the feeding amount monitoring module, and dynamically generates the target operating speed and the target speeds of each component, where the reinforcement learning algorithm takes maximizing the harvesting efficiency and minimizing the grain loss as the optimization objectives, and the Transformer module processes historical operating data and real-time working conditions data to optimize the control strategy; the speed regulation module works in coordination with a PID controller based on a multi-agent reinforcement learning model, and each agent in the multi-agent model corresponds to a reel, a conveyor belt, or a drum, and optimizes the communication between agents through a Transformer to realize the dynamic regulation of the coupling relationship of the speeds of multiple components.

[0021] The system integrates mechanical mechanics sensors (tilted conveyor main shaft torque sensor, drum torque sensor) and optical sensors (lidar, machine vision camera) into the feeding amount monitoring module through a multi-modal sensing fusion architecture, and uses the cross-modal self-attention mechanism of the Transformer architecture to realize the global correlation analysis of time series data and spatial point cloud data; the controller is built-in with a reinforcement learning model to optimize the target operating speed and component speeds with a Markov decision process, and at the same time combines a PID controller to realize closed-loop regulation; the speed regulation module adopts a multi-agent deep deterministic policy gradient algorithm, and each agent independently controls the drive wheel, the reel, and the conveyor belt, and shares the environmental state information through a Transformer encoder to solve the problem of multi-component coupling control.

[0022] The working process of the data fusion unit of the feeding amount monitoring module includes: when the data error of the traditional mechanical sensor exceeds the preset threshold, it automatically switches to the optical sensor data, extracts the lidar point cloud features through a convolutional neural network, and uses the self-attention mechanism of the Transformer to fuse the time series torque data and the spatial point cloud data, and outputs a feeding amount prediction value with an error ≤ 5%.

[0023] The data fusion unit adopts a redundant fault-tolerant design. When the measurement error of traditional mechanical sensors (such as the pressure sensor at the bottom of the conveying trough) exceeds ±10% due to mechanical wear, the system triggers the optical sensing dominant mode: the lidar scans the crop layer at a frequency of 20 Hz to generate point cloud data, extracts geometric features through the PointNet++ network, and fuses them with the YOLOv5 object detection results of machine vision; the Transformer architecture weights different modality data through multi-head self-attention and outputs the predicted value of the feeding amount, with the error rate controlled within ±5% to ensure the monitoring continuity and reliability;

[0024] The system is applicable to the harvesting scenarios of wheat, rice, corn and soybeans, reduces the grain loss rate by 5%-8% by dynamically adjusting the operating speed, and reduces the maintenance cost by 30%-40%.

[0025] It adapts to various crop scenarios through a dynamic adjustment algorithm: for low-stem crops such as wheat and rice, the target feeding amount threshold is set to 8-10 kg / s, and the reference speed of the driving wheel is 3-4 km / h; for high-density crops such as corn and soybeans, the feeding amount threshold is increased to 12-15 kg / s, and the driving wheel speed is reduced to 2-3 km / h; actual tests show that the grain loss rate is reduced from 8.2% of the traditional model to 5.1% (wheat) and from 7.5% to 4.9% (corn), the maintenance frequency is reduced by 42%, and the average annual maintenance cost is reduced by 37%;

[0026] A method for controlling the operating speed of a harvester, including a harvester body 1; a cutter bar 4 is hinged at the front end of the harvester body 1; a reel 3 is installed above the front of the cutter bar 4; driving wheels 2 are symmetrically installed on both sides of the bottom of the harvester body 1; a conveyor belt 5 is arranged behind the cutter bar 4; a threshing device 6 is fixedly connected to the middle of the harvester body 1, and the threshing device 6 is linked with the power output end of the engine through a conveyor belt, and its outlet is connected to the storage bin at the rear of the body; the feeding amount data of traditional sensors and optical sensors are collected in real time, multi-source data are fused through the Transformer architecture and the future feeding amount trend is predicted, the target operating speed and the target rotational speeds of each component are generated based on the reinforcement learning algorithm, single-component closed-loop adjustment is realized through a PID controller, and at the same time, the rotational speed coupling relationship of the driving wheel, the reel and the conveyor belt is coordinated through a multi-agent model, in which the communication logic between multi-agents is optimized by the Transformer to dynamically respond to the change of crop density and ensure the matching of the driving wheel traveling speed, the reel rotational speed and the conveyor belt conveying speed to avoid crop accumulation or incomplete threshing.

[0027] The vehicle body structure adopts a modular design. The driving wheel 2 is connected to the engine output end through a planetary gear reducer, and the torque resolution of the transmission shaft reaches ±5 N·m. The rotating shaft of the reel 3 is rigidly coupled with the hydraulic motor by a flange. The hydraulic system has a closed-loop pressure control (0 - 20 MPa), and the rotational speed adjustment accuracy is ±3 rpm. The surface of the roller of the conveyor belt 5 is coated with a polyurethane anti-slip layer, the chain transmission efficiency is ≥92%, and the motor drive uses a vector control frequency converter with a rotational speed response time ≤0.5 s. The gap between the spike tooth cylinder and the screen of the threshing device 6 is dynamically adjusted by an electric push rod (range 5 - 15 mm) to adapt to the threshing requirements of different crops.

[0028] The cooperative control of the multi-agent model includes: each agent independently controls a transmission component. The driving wheel agent adjusts the rotational speed of the conveyor belt according to the driving speed. The reel agent receives the global operation status information through a Transformer and optimizes the matching relationship between the rotational speed and the driving speed in real time to ensure a stable feeding volume.

[0029] In the multi-agent cooperative control, the driving wheel agent optimizes the driving speed strategy with the Q-learning algorithm, and the state space includes the real-time feeding volume, terrain slope, and engine load. The reel agent uses the Deep Deterministic Policy Gradient (DDPG) algorithm, and the action space is the opening of the hydraulic motor pressure valve (0 - 100%). The conveyor belt agent adjusts the motor rotational speed through PID parameter self-tuning (Ziegler-Nichols method). The Transformer architecture serves as the central coordinator and assigns decision priorities to each agent through attention weights. For example, when the crop density suddenly increases, the weight of the reel agent is increased to 70% to ensure a stable feeding volume.

[0030] The controller is built-in with a fault tolerance mechanism. When any sensor data in the traditional sensor group is abnormal, it automatically switches to the monitoring mode dominated by optical sensors and compensates for data deviation through deep learning algorithms to maintain the continuous operation of the system.

[0031] The fault tolerance mechanism uses a dual-redundant communication bus. When the traditional sensor group fails, the system switches to the optical sensing channel within 50 ms and predicts the missing data through a Long Short-Term Memory (LSTM) network with a prediction error ≤3%. At the same time, the controller activates the safety mode: restricting the maximum speed of the driving wheel to 2 km / h and locking the rotational speed of the reel at 80% of the rated value to avoid equipment overload until manual intervention.

[0032] The PID controller parameters of the rotational speed adjustment module are dynamically optimized through reinforcement learning. Combining the historical operation data of the multi-agent model, the proportional, integral, and differential coefficients are adaptively adjusted to achieve a rotational speed control accuracy of ±5 r / min.

[0033] The parameters of the PID controller are optimized online through deep reinforcement learning. The state inputs include the integral of historical errors, real-time disturbances (such as crop humidity), and component wear coefficients. The dynamic range of the proportional coefficient (Kp) is 0.5 - 2.0, the integral time (Ti) ranges from 1 to 10 seconds, and the derivative time (Td) ranges from 0.1 to 1 second. Experiments show that compared with fixed PID parameters, the dynamic optimization strategy reduces the overshoot of the drive wheel speed by 58%, and the steady-state error of the conveyor belt speed drops from ±8 rpm to ±3 rpm.

[0034] The Transformer architecture adopts a multi-head self-attention mechanism with the number of heads set to 8. The input data includes sensor time-series data, point cloud spatial features, and historical fault records, and the output is the predicted value of the fused feed-in quantity and control instructions.

[0035] The Transformer architecture adopts a pre-training - fine-tuning mode. In the pre-training stage, the mapping relationship between sensor data and feed-in quantity is learned using historical operation data (100,000 groups). In the fine-tuning stage, the attention head weights are optimized by combining online collected data. The input layer contains a 128-dimensional feature vector (64-dimensional time-series sensor data + 64-dimensional point cloud spatial features), and the output layer generates a 32-dimensional control instruction vector (including target speed, component speed, and fault flag bits), with an inference delay ≤20 ms, meeting the real-time control requirements.

[0036] The above has described in detail an embodiment of the present invention, but the content described is only a preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. Any equivalent changes and improvements made according to the scope of the application of the present invention should still fall within the patent coverage scope of the present invention.

Claims

1. A harvester operation speed instruction feedback system, characterized in that It includes a feed quantity monitoring module, a controller and a speed adjustment module; the feed quantity monitoring module adopts multimodal sensor fusion technology, integrates a mechanical sensor group and an optical sensor group, and realizes cross-modal data association analysis through a Transformer architecture; the controller has built-in reinforcement learning algorithm and PID control strategy, receives data from the feed quantity monitoring module and dynamically generates a target operating speed and a target speed of each component, wherein the reinforcement learning algorithm takes maximizing harvesting efficiency and minimizing grain loss as optimization goals, and the Transformer module processes historical operation data and real-time operating condition data to optimize the control strategy; the speed adjustment module works in collaboration with a PID controller based on a multi-agent reinforcement learning model, wherein each agent in the multi-agent model corresponds to a reel, a conveyor belt or a roller, and optimizes communication between agents through Transformer to realize dynamic adjustment of the speed coupling relationship of multiple components.

2. The system according to claim 1, characterized in that The data fusion unit workflow of the feed amount monitoring module includes: when the error of traditional mechanical sensor data exceeds a preset threshold, it automatically switches to optical sensor data, extracts lidar point cloud features through a convolutional neural network, and uses the Transformer's self-attention mechanism to fuse time series torque data and spatial point cloud data, and outputs a feed amount prediction value with an error of ≤5%.

3. The system according to claim 1, characterized in that The system is suitable for wheat, rice, corn and soybean harvesting scenarios. By dynamically adjusting the operating speed, it can reduce the grain loss rate by 5%-8% and reduce maintenance costs by 30%-40%.

4. A method for controlling the operating speed of a harvester based on the system according to any one of claims 1 to 3, characterized in that The invention comprises a harvester body (1); a cutting platform (4) is hingedly connected to the front end of the harvester body (1); a reel (3) is installed on the front upper part of the cutting platform (4); driving wheels (2) are symmetrically installed on both sides of the bottom of the harvester body (1); a conveyor belt (5) is arranged behind the cutting platform (4); a threshing device (6) is fixedly connected to the middle part of the harvester body (1), and the threshing device (6) is linked to the power output end of the engine through the conveyor belt, and its outlet is connected to the storage bin at the rear of the body; the feeding amount data of the traditional sensor and the optical sensor are collected in real time, and the threshing device (6) is connected to the storage bin at the rear of the body through the conveyor belt. The ansformer architecture integrates multi-source data and predicts future feed trends. It generates target operating speeds and target rotation speeds of each component based on a reinforcement learning algorithm, implements closed-loop regulation of a single component through a PID controller, and coordinates the speed coupling relationship between the drive wheel, reel, and conveyor belt through a multi-agent model. The multi-agent optimizes communication logic through Transformer, dynamically responds to changes in crop density, and ensures that the drive wheel speed, reel rotation speed, and conveyor belt speed match each other to avoid crop accumulation or incomplete threshing.

5. The method for controlling the operating speed of a harvester according to claim 4, characterized in that The collaborative control of the multi-agent model includes: each agent independently controls a transmission component, the driving wheel agent adjusts the conveyor belt speed according to the driving speed, and the reel wheel agent receives global operation status information through Transformer, optimizes the matching relationship between the rotation speed and the driving speed in real time, and ensures the stability of the feeding amount.

6. The system according to claim 1, characterized in that The controller has a built-in fault tolerance mechanism. When the data of any sensor in the traditional sensor group is abnormal, it automatically switches to the monitoring mode dominated by the optical sensor, and compensates for the data deviation through a deep learning algorithm to maintain the continuous operation of the system.

7. The system according to claim 1, characterized in that The PID controller parameters of the speed regulation module are dynamically optimized through reinforcement learning, and the proportional, integral and differential coefficients are adaptively adjusted in combination with the historical operation data of the multi-agent model to achieve a speed control accuracy of ±5r / min.

8. The system according to claim 1, characterized in that The Transformer architecture adopts a multi-head self-attention mechanism with 8 heads. The input data includes sensor time series data, point cloud spatial features and historical fault records. The output is the fused feed prediction value and control instructions.

Images