CN108934416A - One kind being based on BP neural network combined harvester multi-parameter operation-control system and method - Google Patents

Abstract

The invention relates to a multi-parameter operation control system and method for a combine harvester based on BP neural network, including a control system and a sensor assembly, the sensor assembly is connected with a stepping motor speed regulating mechanism, a header conveyor, a conveying trough, a driving wheel, The belt pulley is connected to the electric cylinder to obtain the speed signal of the threshing drum, the speed signal of the header conveyor, the speed signal of the conveying chute, the forward speed signal, the speed signal of the cleaning fan and the signal of the concave plate gap. The control system is connected with the electric cylinder. Control the expansion and contraction of the electric cylinder to control the gravure gap. The control system is connected with the stepping motor and the driver, and then controls the hydraulic speed regulating mechanism and the forward speed by controlling the stepping motor and the driver. The invention can realize the real-time monitoring of the six parameters of the rotating speed of the threshing drum, the rotating speed of the conveying trough, the rotating speed of the header screw, the rotating speed of the cleaning fan, the gap between the concave plates and the advancing speed.

Description

A multi-parameter operation control system and method for a combine harvester based on BP neural network

technical field

The invention relates to the technical field of intelligent control of combine harvester operation of agricultural machinery automation, in particular to a multi-parameter operation control system and method for a combine harvester based on BP neural network.

Background technique

At present, because the combine harvester operation control technology mainly adopts fuzzy control technology, the control rules are single, the control parameters are difficult to adjust, and it cannot adapt to the complex and changeable external conditions during field operations. Therefore, a BP neural network-based combine harvester multi-parameter operation control system and method help to solve the above problems.

Foreign developed countries have adopted a variety of sensors to monitor various operating information of the combine harvester, developed an advanced intelligent control system, and improved the intelligence level of the combine harvester. Such as John Deere, Case, New Holland, etc., the electronic monitoring and digital display equipment in the cab of the combine harvester monitors the speed of each operation point, the forward speed of the machine, the height of the stubble, the threshing separation and cleaning loss, and the fullness of the granary. , and automatically adjust the gravure gap, threshing drum speed, fan, auger speed, and forward speed. Foreign researcher Hall determined the setting parameters of the harvester by using data to train the neural network model of the combine harvester to optimize the control system. However, the computer processing speed was limited at that time, and it was not practically applied in the automatic control of the combine harvester. Hiregoudar designed an artificial neural network control algorithm based on the detection of grain moisture content and harvest loss to establish the operating speed control model of the combine harvester. The results of the study confirmed that the properly trained neural network model can be used to predict and judge the amount of post-harvest loss under field conditions. Gundoshmian.TM predicts the operating performance of the combine harvester through the artificial neural network, studies the neural network using the BP error propagation algorithm to establish the operating model of the combine harvester, and obtains the optimal artificial neural network structure through trial and error and trying different structures. Through the network model, the comprehensive performance of wheat yield, crop variety, crop water content, crop height, header height, threshing drum speed, concave plate gap, fan speed, and opening of cleaning sieve was studied.

Compared with the advanced operation control system of the combine harvester of multinational companies in Europe and the United States, the operation process of the domestic combine harvester is mainly based on the manual operation of the machine operator. The automatic operation control system has not been widely used in practice, and many control strategies have been adopted in the research. It is fuzzy control or PID control, which cannot adapt to complex and changeable external conditions during field operations. The overall operation performance of the combine harvester is unstable, frequent blockage faults occur, and the trouble-free working time is short.

Contents of the invention

The purpose of the present invention is to provide a multi-parameter operation control system and method for a combine harvester based on BP neural network, which can realize the multi-parameter monitoring of the combine harvester and the requirements for automatic adjustment of the forward speed and the gap between the concave plates, so as to improve the combined harvest. The purpose of machine operation performance and work efficiency.

To achieve the above object, the present invention adopts the following technical solutions:

A multi-parameter operation control system for a combine harvester based on BP neural network, including a control system and a sensor assembly, the sensor assembly is connected with a stepping motor speed regulating mechanism, a header conveyor, a conveying trough, a driving wheel, a pulley, and an electric cylinder , to obtain the threshing drum speed signal, the header conveyor speed signal, the conveying chute speed signal, the forward speed signal, the cleaning fan speed signal and the concave plate gap signal respectively. The control system is connected with the electric cylinder, and by controlling the telescopic movement of the electric cylinder To control the gravure gap, the control system is connected with the stepping motor and the driver, and then controls the hydraulic speed regulating mechanism and the forward speed by controlling the stepping motor and the driver.

Further, it also includes a touch screen and an alarm connected to the control system.

A method for controlling multi-parameter operation of a combine harvester based on BP neural network, comprising the following steps:

(1) Adjust the concave plate gap through the electric cylinder to obtain multiple sets of parameter data for the concave plate gap;

(2) After making a difference between the parameter data and the corresponding setting value of the control system, the total sample data is obtained, and the total sample data is normalized;

(3) In the normalized data of the total sample data, from the initial moment, one piece of data is selected every 1 second to form multiple sets of data as network training sample data, and the remaining sample data are randomly divided into test samples data and test sample data;

(4) Calculate the error δ _p (t) of the training network when the p-th sample data is adjusted for t times of weight, if the total error Then carry out the calculation of test sample data error ε ₁ and test sample data error ε ₂ , if ε ₁ , ε ₂ ∈ [0,1.2ε], then output the weight coefficient matrix, and the learning is over, otherwise according to δ(t), press Gradient descent method reverse calculation, adjust the weight coefficient.

Calculate the error δ _p (t) of the training network when the p-th sample data is adjusted for t times of weight, and it is calculated by the following formula:

Among them, δ _p (t) represents the error of the training network when the p-th sample data undergoes t-time weight adjustments, and w _ij (t) represents the weight coefficient of the j-th input of node i after t-time weight adjustments; I _jp represents the jth input of node i, and y _kp represents the kth data of the pth sample output data.

It can be seen from the above technical scheme that the multi-parameter operation control system and method based on BP neural network combine harvester described in the present invention can realize the control of the speed of the threshing drum, the speed of the conveying trough, the speed of the header screw, the speed of the cleaning fan, the concave Real-time monitoring of 6 parameters of board clearance and forward speed; the system can record and collect input data (rotation speed of threshing drum, header conveyor and conveying chute) and output data (advance speed, concave board clearance). The network learning algorithm enables the forward speed of the combine harvester and the gap between the concave plates to be automatically adjusted according to the control quantity output by the neural network, making the system easy to operate, safe and reliable, intuitive interface, and high operating efficiency.

Description of drawings

Fig. 1 is a system diagram of the present invention;

Fig. 2 is a schematic structural diagram of a stepping motor speed regulating mechanism of the present invention;

Fig. 3 is a schematic structural diagram of the concave plate gap adjustment mechanism of the present invention;

Fig. 4 is a schematic diagram of the BP neural network structure of the combine harvester operation control system of the present invention;

Fig. 5 is a flow chart of the method of the present invention.

Detailed ways

The present invention will be further described below in conjunction with accompanying drawing:

As shown in Figure 1, the combined harvester multi-parameter operation control system based on BP neural network of the present embodiment includes control system and sensor assembly, touch screen and alarm, sensor assembly and stepper motor speed regulating mechanism, header conveyor, Conveying trough, driving wheel, belt pulley and electric cylinder are connected to obtain the threshing drum speed signal, header conveyor speed signal, conveying trough speed signal, forward speed signal, cleaning fan speed signal and concave plate gap signal respectively, and the control system and electric motor Cylinder connection, control the gravure gap by controlling the expansion and contraction of the electric cylinder, the control system is connected with the stepping motor and the driver, and then controls the hydraulic speed regulating mechanism and the forward speed by controlling the stepping motor and the driver, the touch screen and the alarm and the control system connected.

As shown in Figure 2, Figure 2 is a schematic diagram of the stepping motor speed regulating mechanism, which is composed of the main speed change handle 1, the transition rod 2, the stepping motor 3, the transition sleeve 4, the HST connecting rod 5 and the HST device 6. Under the action of the control signal, the motor 3 rotates at a certain angle in a certain direction at a certain angular speed, drives the transition sleeve 4 and the HST connecting rod 5 to move, and pushes the HST device 6, thereby achieving the goal of adjusting the forward speed of the combine harvester and the speed of other working parts. Purpose.

As shown in Figure 3, the accompanying drawing 3 is a schematic diagram of the structure of the concave plate gap adjustment mechanism. The structure of the concave plate gap adjustment mechanism is mainly composed of three parts: The angle between the symmetrical planes of the plate sieve is 20°. The concave plate screen 3 adjusts the size of the gap between the threshing drum 1 and the threshing drum 1 through the joint action of 6 electric cylinders 2 . The movement process of the 6 electric cylinders is consistent under the control of the system, and the adjustment amount and direction are the same.

The six parameters of threshing drum speed, conveying trough speed, header screw speed, concave plate gap, forward speed and cleaning fan speed are monitored in real time through sensor components, which is convenient for the machine operator to perform manual operation; The real-time data of the three parameters of trough speed and header screw speed are input into the neural network model after training to output the control amount to the stepping motor and electric cylinder, and the purpose of adjusting the forward speed of the combine harvester and the gap between the concave plates to obtain the best operating performance; The function is to complete the neural network training according to the BP learning algorithm and determine the weight coefficient matrix of the network.

The sensor component collects the data of 5 parameters of threshing drum speed, conveyor trough speed, header screw speed, concave plate gap and forward speed, and sends the data to the control system. The data changes of the three input parameters of the helical speed of the cutting table. After training, the neural network outputs the control amount to the stepping motor and electric cylinder to adjust the forward speed and the size of the concave plate gap. During normal automatic operation, each parameter changes within the allowable range , the background color of the corresponding display box is green; when a certain parameter exceeds the allowable value range, the background color of the corresponding display box will be highlighted in red and strobe. , to prevent damage to machine parts.

The control system needs to use the BP neural network learning algorithm to complete the control of the stepper motor, the electric cylinder, the adjustment of the forward speed, and the gap between the concave plates. Attached figure 4 is a schematic diagram of the BP neural network structure of the combine harvester operation control system. As shown in the figure, this The network adopts a 3-5-2 three-layer network structure, that is, the input layer has 3 input parameters (the speed change of the threshing drum, the speed change of the conveying trough, the speed change of the header screw conveyor), and the middle layer has 5 neuron nodes, The output layer has 2 output parameters (concave gap change, forward speed change).

The control system needs to learn the algorithm steps through the BP neural network, as shown in Figure 5. The entire algorithm process includes two parts, one is the sample data processing and preparation stage; the other is the BP neural network learning stage. The sample data processing and preparation stage mainly includes importing total sample data, data normalization processing, dividing training sample data, testing sample data and testing sample data; the BP neural network learning stage mainly includes the following steps:

(1) Before the machine operation, adjust the concave plate gap, and perform manual effective operation for 5 minutes when the concave plate gap is 15mm, 20mm and 25mm respectively. The equipment records the monitoring parameter data every 0.5s, and obtains a total of concave plate gaps of 15mm and 20mm 600 pieces of recorded parameter data for each of the 3 groups of 25mm and 25mm, the total sample data is 1800 pieces after making a difference with the corresponding setting value of the control system, and normalized;

(2) Among the three sets of normalized data corresponding to the concave plate gaps of 15mm, 20mm and 25mm, from the initial moment, one piece of data is selected every 1 second, and a total of 600 pieces of data are formed as network training sample data; The remaining 1200 pieces of data are randomly divided into 600 pieces of test sample data and 600 pieces of test sample data;

(3) The network parameters are initialized, and the network learning rate is determined to be 0.05, the error precision is 0.001, the action function adopts an asymmetric Sigmoid function, and the initial weight coefficient matrix is a random non-zero matrix.

(4) According to the BP learning algorithm, calculate the error δ _p (t) of the training network when the p-th sample data is adjusted for t times of weight. Suppose the input data of the p-th sample: x _p =(x _1p ,x _2p ,x _3p ); the output data of the sample: y _p =(y _1p ,y _2p ). but

Among them, w _ij (t) is used for the weight coefficient of the jth input of node i after t times of weight adjustment; I _jp is the jth input of node i.

(5) Calculate the total error of the training sample data δ(t)=∑δ _p (t); if δ(t)=∑δ _p (t)≤ε, the test sample data error ε ₁ and the test sample data error ε ₂ calculation; if δ(t)=∑δ _p (t)>ε, if ε ₁ ,ε ₂ ∈[0,1.2ε], then output the weight coefficient matrix, and the learning is over; otherwise, skip to step (6 ).

(6) According to δ(t), reverse calculation according to the gradient descent method, and adjust the weight coefficient:

When the control system of the combine harvester passes the data collected manually, the input and output sample data are formed through data processing, and input to the control system BP neural network learning algorithm program to determine the weight coefficients of the neural network connection, and the control system completes the sample data learning After training, the combine harvester can enter the automatic operation mode; when it enters the automatic operation mode, the control system will obtain the speed data of the threshing drum, the speed of the conveying trough, and the speed of the header screw conveyor through the adjacent interruption time difference according to the sensor components, and then through data processing The program obtains the change data of each parameter, and then forms the input parameter data through data normalization processing, enters the trained BP neural network control program to obtain the corresponding concave plate gap change and forward speed change, forms an output control signal, and then inputs To the pulse end and direction end of the driver in the electric cylinder and the stepper motor driver, the concave plate gap between the concave plate sieve and the threshing drum (attachment 3) and the rotation angle of the stepper motor (attachment 2) have been adjusted separately the goal of.

The above-mentioned embodiments are only descriptions of preferred implementations of the present invention, and are not intended to limit the scope of the present invention. Variations and improvements should fall within the scope of protection defined by the claims of the present invention.

Claims (4)

1. one kind is based on BP neural network combined harvester multi-parameter operation-control system, it is characterised in that: including control system And sensor module, the sensor module and stepper motor speed adjusting gear, ceding of Taiwan conveyer, delivery chute, driving wheel, belt pulley And electric cylinder connection, threshing cylinder tach signal is obtained respectively, ceding of Taiwan conveyer tach signal, delivery chute tach signal, is advanced Speed signal, cleaning fan tach signal and concave clearance signal, the control system are connect with electric cylinder, electronic by controlling The extension and contraction control intaglio plate gap of cylinder, the control system are simultaneously connect with stepper motor and driver, pass through control stepper motor and drive Dynamic device and then control hydraulic speed regulating mechanism and forward speed.

2. according to claim 1 be based on BP neural network combined harvester multi-parameter operation-control system, feature exists In: it further include the touch screen and alarm being connect with control system.

3. one kind is based on BP neural network combined harvester multi-parameter job control method, which comprises the following steps:

(1) concave clearance is adjusted by electric cylinder, obtains concave clearance multiple groups supplemental characteristic；

(2) supplemental characteristic and control system are correspondingly arranged value to make to obtain total number of samples evidence after difference, and by total number of samples According to being normalized；

(3) since initial time, 1 data was selected every 1 second according to being normalized in data in total number of samples, formed altogether Multi-group data is randomly divided into test samples data and test sample as training sample data, and by remaining sample data Data；

(4) error delta of the pth sample data through training network when t weighed value adjusting is calculated_p(t), if overall errorSample data of then testing error ε₁With test sample data error ε₂Calculating, if ε₁,ε₂∈ [0,1.2 ε], then export weight coefficient matrix, and study terminates, otherwise according to δ (t), by gradient descent method retrospectively calculate, adjustment power Value coefficient.

4. according to claim 3 be based on BP neural network combined harvester multi-parameter job control method, feature exists In: the error delta of training network when the calculating pth sample data is through t weighed value adjusting_p(t), it is calculated by the following formula It arrives:

Wherein, δ_p(t) error of the pth sample data through training network when t weighed value adjusting, w are indicated_ij(t) it indicates to weigh through t times The weight coefficient of j-th of node i input when value adjustment；I_jpIndicate j-th of node i input, y_kpIndicate that pth bar sample exports number K-th of data in.