Intelligent control method of wheat combine harvester

Abstract

According to the intelligent control method of the wheat combine harvester, sensor detection data is used as input, an operation quality prediction value of a period of time in the future is obtained through an LSTM neural network prediction model, and therefore harvester operation performance parameters such as the impurity rate, the loss rate, the breakage rate and the yield of grains are used as evaluation indexes; simulation and prediction results are made through historical time sequence data, a control instruction is generated according to predicted operation performance parameters and parameter thresholds of the operation performance parameters, and an optimal control decision result for an expected state of an actuator is output, so that the accuracy of parameter prediction of the combine harvester is remarkably improved; more accurate control parameters of the combine harvester can be obtained, and the requirement for automatic intelligent control of the combine harvester is met.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for combine harvesters, and in particular to an intelligent control method for a wheat combine harvester. Background Technology

[0002] Loss rate and impurity content are key indicators for evaluating the working quality of wheat harvesters. Harvester operators relying on experience to adjust harvesting parameters suffer from decision-making delays and inaccurate control, making it difficult to meet the production demands of real-time decision-making and control. New-generation electronic information technology provides a technological foundation for intelligent operation in the wheat harvesting process. For example, the Internet of Things (IoT) enables information interaction between users (people), harvesters (machines), sensors (objects), and systems (cloud), further enabling real-time interoperability between harvester operation data, models, and the harvester itself. However, existing technologies still have shortcomings in the methods for parameter prediction and automatic control of combine harvesters. The accuracy of predictions is low, resulting in insufficient precision in setting control parameters for combine harvesters, failing to meet the requirements for automatic intelligent control of combine harvesters. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent control method for wheat combine harvesters, which improves the accuracy of combine harvester parameter prediction and obtains more precise combine harvester control parameters.

[0004] The technical solution adopted by this invention to solve the above-mentioned technical problems is: an intelligent control method for a wheat combine harvester, comprising the following steps: S1. Multiple sensors are installed on the wheat combine harvester, including header speed sensor, drum speed sensor, fan speed sensor, grain yield sensor, grain loss sensor and grain impurity sensor, to detect header speed, drum speed, fan speed, grain yield, grain loss and grain impurity data during the operation of the wheat combine harvester. S2. The raw data collected by the sensor is filtered using a Gaussian weighted moving average filtering algorithm. First, the one-dimensional form of the Gaussian function is calculated as the weight value of the signal: ; Then, normalization is performed: ; Noise is eliminated by weighted averaging of the signal, and the output filtered result y is: ; In the above formula, x is the distance from the position within the window to the center of the window; σ is the standard deviation of the function; G(x) represents the Gaussian weight value corresponding to the signal at a distance x from the center of the window within the filtering window; w iThis represents the normalized weight value corresponding to the signal at the i-th position within the filtering window; N represents the window length. S3. The filtered data is fed into the LSTM neural network prediction model. The prediction model preprocesses the data obtained over a past period and then feeds the preprocessed historical time series data into the model. As input to the input layer, the output is the hidden state at each time step. and the output of the previous layer As the input to the next layer, the output of the last layer is ,Will Mapping to the predicted output dimension Calculate the predicted value of future work quality. : ; In the above formula, W out h is the weight matrix of the output layer of the LSTM neural network. T b represents the hidden state of the LSTM neural network at the last time step T; out This is the bias vector for the output layer of the LSTM neural network, used to adjust the baseline of the prediction results; The LSTM neural network includes an input layer, multiple hidden layers, and an output layer. The basic unit is a memory cell. Three gating mechanisms control the forgetting, input, and output of memory information and update the memory cells. The forget gate integrates the current input value and the output value from the previous time step, and then compresses the input to the interval (0, 1) using the sigmoid function. If the vector becomes 0 after passing through the sigmoid layer, the information of that component is forgotten. The input gate extracts valid information from the vector using the tanh function, and then controls whether the memory cell enters the cell state using the sigmoid function; The output gate integrates the current input value with the output value of the previous time step, then extracts information through the sigmoid function, and then compresses and maps the current cell state to the interval (-1, 1) through the tanh function to obtain the output value at the current time step; S4. Input the predicted value of future operation quality into the control decision algorithm to obtain the initial control strategy; The control decision algorithm is a particle swarm optimization algorithm, and the objective function is expressed as: ; In the above formula, This represents the actual operational performance of the wheat combine harvester at time i, including grain loss and grain impurity content. The target job performance is adjusted based on the predicted values ​​of future job quality, where N is the number of data points; Particle swarm optimization (PSO) algorithms adjust the positions of particles by solving the objective function in the solution space, thereby obtaining the optimal control parameters. S5, the wheat combine harvester continues to operate according to the optimal control parameters.

[0005] Preferably, the raw data in S1 is transmitted to the edge computing device via the CAN bus. The edge computing device executes S2 to transmit the filtered data to the cloud server. The cloud server executes S3 and S4, and then sends the optimal control parameters to the edge computing device via the network communication unit. The edge computing device receives and parses the decision instructions, and sends the control instructions to the corresponding actuators via the CAN bus. The actuators receive the control instructions from the CAN bus and execute them.

[0006] Preferably, the input gate is defined as i in time step t. t The forget gate is f t The output gate is 0 t Memory cells are updated to C t The three gating and memory cell update formulas are as follows: ; ; ; ; ; In the above formula, h t-1 x represents the hidden state of the LSTM neural network in the previous time step. t This represents the original feature vector of the LSTM neural network at the current time step; [h] t-1 , x t ] represents the vector concatenation operation, used as input for gating computation; tanh represents the hyperbolic tangent activation function; W i W c W f W o These are the learnable weight matrices for the input gate, candidate cell state, forget gate, and output gate, respectively; b i b c b f b o These are the learnable bias vectors corresponding to the input gate, candidate cell state, forget gate, and output gate, respectively.

[0007] According to the above technical solution, the beneficial effects of the present invention are: This invention uses sensor detection data as input and an LSTM neural network prediction model to obtain predicted values ​​of operation quality for a future period. It then uses harvester performance parameters such as grain impurity rate, loss rate, breakage rate, and yield as evaluation indicators. Simulation and prediction results are generated using historical time-series data. Based on the predicted operation performance parameters and their threshold values, control commands are generated, and the optimal control decision for the desired actuator state is output. This significantly improves the accuracy of combine harvester parameter prediction, enabling more precise combine harvester control parameters and meeting the requirements for automatic intelligent control of combine harvesters. Attached Figure Description

[0008] Figure 1 This is a block diagram of the control system module for an embodiment; Figure 2 This is a schematic diagram of an LSTM neural network prediction model. Detailed Implementation

[0009] A method for intelligent control of a wheat combine harvester includes the following steps: S1. Multiple sensors are installed on the wheat combine harvester, including header speed sensor, drum speed sensor, fan speed sensor, grain yield sensor, grain loss sensor and grain impurity sensor.

[0010] The aforementioned sensors are used to detect data on header speed, drum speed, fan speed, grain yield, grain loss, and grain impurity content during the operation of the wheat combine harvester.

[0011] S2. The raw data collected by the sensor is filtered using a Gaussian weighted moving average filtering algorithm. The Gaussian weighted moving average filtering algorithm eliminates noise by weighting the signal. The weights are calculated based on a Gaussian distribution, with larger weights closer to the center point and smaller weights farther from the center point.

[0012] First, calculate the one-dimensional form of the Gaussian function as the weight value of the signal: ; Then, normalization is performed: ; Noise is eliminated by weighted averaging of the signal, and the output filtered result y is: .

[0013] In the above formula, x is the distance from the position within the window to the center of the window; σ is the standard deviation of the function; G(x) represents the Gaussian weight value corresponding to the signal at a distance x from the center of the window within the filtering window; w iThis represents the normalized weight value corresponding to the signal at the i-th position within the filtering window; N represents the window length.

[0014] S3. The filtered data is fed into the LSTM neural network prediction model. The prediction model preprocesses the data obtained over a past period and then feeds the preprocessed historical time series data into the model. As input to the input layer, the output is the hidden state at each time step. and the output of the previous layer As the input to the next layer, the output of the last layer is ,Will Mapping to the predicted output dimension Calculate the predicted value of future work quality. : ; In the above formula, W out h is the weight matrix of the output layer of the LSTM neural network. T b represents the hidden state of the LSTM neural network at the last time step T; out This is the bias vector of the output layer of the LSTM neural network, used to adjust the baseline of the prediction results.

[0015] like Figure 2 As shown, the LSTM neural network includes an input layer, multiple hidden layers, and an output layer. The basic unit is the memory cell. The forgetting, input, and output of memory information are controlled by three gating mechanisms, and the memory cell is updated.

[0016] The forget gate determines which memory cells to forget based on each input and the previous output. The forget gate first integrates the current input value and the previous output value, and then compresses the input to the interval (0, 1) using the sigmoid function. If the vector becomes 0 after passing through the sigmoid layer, the information of that component is forgotten.

[0017] The input gate controls whether the data at time t is incorporated into the control unit of the cell state, that is, it extracts the effective information from the input. The input gate first extracts the effective information from the vector through the tanh function, and then controls whether the memory cell enters the cell state through the sigmoid function.

[0018] The output gate is a neural layer used to calculate the output value at the current time step. The output gate first integrates the current input value with the output value at the previous time step, then extracts information through the sigmoid function, and finally compresses and maps the current cell state to the interval (-1, 1) through the tanh function to obtain the output value at the current time step.

[0019] The input gate is defined as i in time step t. t The forget gate is ft The output gate is 0 t Memory cells are updated to C t The three gating and memory cell update formulas are as follows: ; ; ; ; .

[0020] In the above formula, h t-1 x represents the hidden state of the LSTM neural network in the previous time step. t This represents the original feature vector of the LSTM neural network at the current time step; [h] t-1 , x t ] represents the vector concatenation operation, used as input for gating computation; tanh represents the hyperbolic tangent activation function; W i W c W f W o These are the learnable weight matrices for the input gate, candidate cell state, forget gate, and output gate, respectively; b i b c b f b o These are the learnable bias vectors corresponding to the input gate, candidate cell state, forget gate, and output gate, respectively.

[0021] S4. Input the predicted value of future operation quality into the control decision algorithm to obtain the initial control strategy.

[0022] The control decision algorithm is a particle swarm optimization algorithm, and the objective function is expressed as: ; In the above formula, This represents the actual operational performance of the wheat combine harvester at time i, including grain loss and grain impurity content. N represents the target job performance after adjustments based on predicted future job quality values, where N is the number of data points.

[0023] Particle swarm optimization (PSO) algorithms adjust the positions of particles by solving the objective function in the solution space, thereby obtaining optimal control parameters and decision results, including concave plate gap, cutting table height, screen opening, drum speed, fan speed, and forward speed.

[0024] S5, the wheat combine harvester continues to operate according to the optimal control parameters.

[0025] like Figure 1 As shown in the embodiment, a control system applying this intelligent control method is also provided, which, in addition to the aforementioned sensors, also includes a cloud server, an in-vehicle edge computing device, a data acquisition unit, a controller, and an actuator.

[0026] The data acquisition unit is responsible for collecting detection data from the sensors and connecting it to the CAN bus.

[0027] The controller is responsible for reading control commands from the CAN bus and sending them to the actuators.

[0028] The actuator includes a travel motor and a push rod motor, used to control the adjustment of parameters such as the gap between the concave plates and the adjustment of the working speed. The actuators that can be connected include concave plate gap adjustment devices, cutting table height adjustment devices, screen opening adjustment devices, drum speed adjustment devices, fan speed adjustment devices, forward speed adjustment devices, etc.

[0029] Edge computing devices receive, process, and send communication signals via a CAN bus. They consist of four modules: a data storage unit, a data processing unit, an algorithm decision-making unit, and a network communication unit. These modules are used to establish communication, data processing, data storage, and control algorithm operation between the combine harvester and the cloud server.

[0030] The data storage unit and data processing unit are used to process the sensor data.

[0031] The algorithm decision unit is used to run the control algorithm to obtain control commands, which are then sent to the actuators via the CAN bus for specific execution.

[0032] The network communication unit is used to establish a network connection with the cloud server, and is responsible for sending processed sensor data to the cloud server and receiving decision instructions from the cloud server.

[0033] The cloud server is the human-computer interaction module, which includes a web visualization page and a simulation prediction function for job quality. The web visualization page contains a real-time mapping of the combine harvester and its key components in virtual space, restoring the grain combine harvester and its operation process from four dimensions: geometry, physics, behavior, and rules.

[0034] The cloud server collects real data from the grain combine harvester, keeps it synchronized with both the physical and virtual harvesters, and further analyzes and predicts the attributes and status of the harvester in the next step based on the physical mechanisms and rules of the harvester operation.

[0035] The web visualization page consists of a front-end page and a back-end executable program. The front-end page displays the combine harvester's operation process and real-time operating data, while the back-end executable program receives and processes data from the front-end and executes simulation prediction strategies.

[0036] The raw data from S1 is transmitted to the edge computing device via the CAN bus. The edge computing device executes S2 to transmit the filtered data to the cloud server. The cloud server executes S3 and S4, and then sends the optimal control parameters to the edge computing device through the network communication unit.

[0037] Edge computing devices receive and parse decision commands, then send control commands to the corresponding actuators via the CAN bus. The actuators receive the control commands from the CAN bus and execute them. Sensors update their detection data and upload the information.

Claims

1. An intelligent control method for a wheat combine harvester, characterized in that, Includes the following steps: S1. Multiple sensors are installed on the wheat combine harvester, including header speed sensor, drum speed sensor, fan speed sensor, grain yield sensor, grain loss sensor and grain impurity sensor, to detect header speed, drum speed, fan speed, grain yield, grain loss and grain impurity data during the operation of the wheat combine harvester. S2. The raw data collected by the sensor is filtered using a Gaussian weighted moving average filtering algorithm. First, the one-dimensional form of the Gaussian function is calculated as the weight value of the signal: ； Then, normalization is performed: ； Noise is eliminated by weighted averaging of the signal, and the output filtered result y is: ； In the above formula, x is the distance from the position within the window to the center of the window; σ is the standard deviation of the function; G(x) represents the Gaussian weight value corresponding to the signal at a distance x from the center of the window within the filtering window; w i This represents the normalized weight value corresponding to the signal at the i-th position within the filtering window; N represents the window length. S3. The filtered data is fed into the LSTM neural network prediction model. The prediction model preprocesses the data obtained over a period of time and then feeds the preprocessed historical time series data into the model. As input to the input layer, the output is the hidden state at each time step. and the output of the previous layer As the input to the next layer, the output of the last layer is ,Will Mapping to the predicted output dimension Calculate the predicted value of future work quality : ； In the above formula, W out h is the weight matrix of the output layer of the LSTM neural network. T b represents the hidden state of the LSTM neural network at the last time step T; out This is the bias vector for the output layer of the LSTM neural network, used to adjust the baseline of the prediction results; The LSTM neural network includes an input layer, multiple hidden layers, and an output layer. The basic unit is a memory cell. Three gating mechanisms control the forgetting, input, and output of memory information and update the memory cells. The forget gate integrates the current input value and the output value from the previous time step, and then compresses the input to the interval (0, 1) using the sigmoid function. If the vector becomes 0 after passing through the sigmoid layer, the information of that component is forgotten. The input gate extracts valid information from the vector using the tanh function, and then controls whether the memory cell enters the cell state using the sigmoid function; The output gate integrates the current input value with the output value of the previous time step, then extracts information through the sigmoid function, and then compresses and maps the current cell state to the interval (-1, 1) through the tanh function to obtain the output value at the current time step; S4. Input the predicted value of future operation quality into the control decision algorithm to obtain the initial control strategy; The control decision algorithm is a particle swarm optimization algorithm, and the objective function is expressed as: ； In the above formula, This represents the actual operational performance of the wheat combine harvester at time i, including grain loss and grain impurity content. The target job performance is adjusted based on the predicted values ​​of future job quality, where N is the number of data points; Particle swarm optimization (PSO) algorithms adjust the positions of particles by solving the objective function in the solution space, thereby obtaining the optimal control parameters. S5, the wheat combine harvester continues to operate according to the optimal control parameters.

2. The intelligent control method for a wheat combine harvester according to claim 1, characterized in that: The raw data in S1 is transmitted to the edge computing device via the CAN bus. The edge computing device executes S2, transmitting the filtered data to the cloud server. The cloud server executes S3 and S4, and then sends the optimal control parameters to the edge computing device via the network communication unit. The edge computing device receives and parses the decision instructions, and sends the control instructions to the corresponding actuators via the CAN bus. The actuators receive the control instructions from the CAN bus and execute them.

3. The intelligent control method for a wheat combine harvester according to claim 1, characterized in that: The input gate is defined as i in time step t. t The forget gate is f t The output gate is 0 t Memory cells are updated to C t The three gating and memory cell update formulas are as follows: ； ； ； ； ； In the above formula, h t-1 x represents the hidden state of the LSTM neural network in the previous time step. t This represents the original feature vector of the LSTM neural network at the current time step; [h] t-1 , x t ] represents the vector concatenation operation, used as input for gating computation; tanh represents the hyperbolic tangent activation function; W i W c W f W o These are the learnable weight matrices corresponding to the input gate, candidate cell state, forget gate, and output gate, respectively; b i b c b f b o These are the learnable bias vectors corresponding to the input gate, candidate cell state, forget gate, and output gate, respectively.

Images