Rice and wheat combine harvester cleaning loss prediction model based on dynamic Bayesian

A dynamic Bayesian network model predicts cleaning loss rates in rice and wheat combine harvesters by analyzing the influence of wind distributor plate angle, fish scale screen opening, and fan speed, achieving high accuracy in predicting cleaning loss trends.

CN119089137APending Publication Date: 2024-12-06SUZHOU ZHISHOU AGRICULTURAL TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411141573.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-20
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Existing technologies face challenges in accurately predicting the cleaning loss rate of rice and wheat combine harvesters, which is crucial for assessing their operating performance.

Method used

A dynamic Bayesian network model is developed to predict cleaning loss rate by considering the influence of cleaning operation parameters such as the angle of the wind distributor plate, the opening of the fish scale screen, and the fan speed, incorporating causal relationships between these factors and their impact on cleaning loss rates at adjacent times.

Benefits of technology

The model achieves high accuracy in predicting the changing trend of cleaning loss rates, with training and test set accuracies of 84.43% and 83.33%, respectively, by effectively modeling the dynamic interactions of these parameters.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089137A_ABST
    Figure CN119089137A_ABST
Patent Text Reader

Abstract

The invention relates to the field of rice and wheat combine harvester cleaning loss rate prediction, in particular to a rice and wheat combine harvester cleaning loss prediction method based on a dynamic Bayesian rice and wheat combine harvester cleaning loss prediction model, and cleaning operation parameters which directly influence the cleaning loss rate at the same moment are the air distribution plate angle, the scale sieve opening degree and the fan rotating speed respectively. Relevance exists between the cleaning losses at adjacent moments, and the following causal relationship is defined: Q (QtQt-1, Xt, Yt, Zt); in the formula, Qt is the cleaning loss amount, and subscripts t-1 and t respectively represent the current moment and the prediction moment; the number of the states of all variables in the formula is three, and the dynamic Bayesian network is expressed to be expanded to different time sequences. The rice and wheat combine harvester cleaning loss rate prediction model based on the dynamic Bayesian network is established by judging the influence of cleaning operation parameters on the cleaning loss rate and the influence between the cleaning loss rates in adjacent time, and the rice and wheat combine harvester cleaning loss rate prediction model is obtained through experimental data and result analysis. It can be known that the prediction model has high accuracy in predicting the change trend of the cleaning loss.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the field of prediction of cleaning loss rate of rice and wheat combine harvesters, and in particular to a dynamic Bayesian prediction model for cleaning loss of rice and wheat combine harvesters. Background Art

[0002] With the development of intelligent sensing technology and intelligent control technology, the popularity of rice and wheat combine harvesters in my country and the continuous improvement of mechanization level, the automation and intelligence of machinery have become a major trend in the development of machinery. The cleaning device is one of the key devices of the rice and wheat combine harvester, and the cleaning loss rate is one of the important indicators for measuring the operating performance of the intelligent rice and wheat combine harvester. Therefore, domestic and foreign scholars have conducted a lot of research on the intelligent control of the operating parameters of the cleaning device and the monitoring of operating performance. Predicting the cleaning loss rate through the working conditions of the cleaning operating parameters is a powerful way to solve the problem.

[0003] Qiao Guochun conducted a rice cleaning experiment on a rice cleaning test bench using fan speed, fan inclination angle and crank speed as experimental factors and loss rate as evaluation index. He obtained the regression models of fan speed, fan inclination angle and crank speed and cleaning loss rate respectively, and used the contribution efficiency method to obtain the main factors affecting the cleaning loss rate.

[0004] Li Yaoming and others conducted airflow field measurement tests and rice cleaning tests on a self-developed cleaning test bench. Using neural network technology, the wind-screen cleaning airflow field was studied, and three BP neural network models were established: the relationship between two wind factors (centrifugal fan speed and air outlet angle) and the distribution of the cleaning airflow field, the relationship between the distribution of the cleaning airflow field and the cleaning effect, and the relationship between the two wind factors and the cleaning effect. The prediction test was carried out using experimental data, and the prediction results proved the effectiveness of the network model.

[0005] At present, how to improve the accuracy of predicting the cleaning loss rate is still a research focus. Based on this, the present invention provides a dynamic Bayesian rice-wheat combine harvester cleaning loss prediction model with high accuracy. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides a rice-wheat combine harvester cleaning loss prediction model based on dynamic Bayesian. The cleaning operation parameters that directly affect the cleaning loss rate at the same time are the wind plate angle, the fish scale screen opening and the fan speed. There is a correlation between the cleaning losses at adjacent times, and the following causal relationship is defined:

[0007] Q(Q t |Q t-1 , X t , Y t , Z t )

[0008] Where: Q t To clear the loss amount, the subscripts t-1 and t represent the current time and the predicted time respectively; the states of all variables in the formula are three, and the dynamic Bayesian network is represented as extended to different time series.

[0009] Preferably, the establishment of the cleaning loss prediction model is divided into the following steps:

[0010] (1) Select variables

[0011] Select a set of time-related variables and use them to represent a set of physical data values. Indicates the physical data value of X in state i at time t, and divides the time slice into time t, t-1, and t-2 according to △t=1;

[0012] Select parameters X, Y, and Z to form a variable set that affects the cleaning loss rate, where X represents the angle of the air distributor, Y represents the opening of the fish scale screen, and Z represents the fan speed. The value range of each variable must consist of a series of discrete values.

[0013] (2) Definition of causality

[0014] (2-1) The effect of the angle of the wind splitter on the cleaning loss rate is described as a network of causal relationships between variables: represents the state i of the wind distributor angle at time t, Indicates the state i of the wind distributor angle at time t-1. There are two time slices of wind distributor angles, using the probability table The causal relationship of the effect of the wind distributor angle X on the cleaning loss rate is shown;

[0015] (2-2) The effect of fish scale screen opening on the cleaning loss rate is described as a network of causal relationships between variables. Indicates the state i of the fish scale screen opening at time t, Indicates the state i of the fish scale screen opening at time t-1. There are two time slices of fish scale screen opening, using the probability table It represents the causal relationship of the effect of fish scale screen opening Y on the cleaning loss rate;

[0016] (2-3) The effect of fan speed on the cleaning loss rate is described as a network of causal relationships between variables. represents the state i of the fan speed at time t, Indicates the state i of the fan speed at time t-1. There are two time slices of fan speed, using the probability table The causal relationship of the fan speed Z on the cleaning loss rate is shown;

[0017] (3) Experiment and analysis

[0018] (3-1) Experimental data processing:

[0019] (3-2) Construction of dynamic Bayesian network model

[0020] By assigning the following relationship to variables, it becomes a dynamic Bayesian network model, function DBN(), adding variables to the DBN function to automatically generate a dynamic Bayesian model;

[0021] The function dbn.fit() is used to automatically calculate the probability in dynamic Bayesian. The training set is input into the dbn.fit() function to automatically calculate the probability of X, Y, and Z being in state 0, state 1, or state 2.

[0022] 1. variables=

[0023] 2.(('X', 0), ('Q', 0));

[0024] 3.(('Y', 0), ('Q', 0));

[0025] 4.(('Z', 0), ('Q', 0));

[0026] 6.(('X',0),('X',1));

[0027] 7.(('Y',0),('Y',1));

[0028] 8.(('Z',0),('Z',1));

[0029] 10.(('Q', 0), ('Q', 1));

[0030] 11.(('X', 1), ('Q', 1));

[0031] 12.(('Y', 1), ('Q', 1));

[0032] 13.(('Z', 1), ('Q', 1));

[0033] 15.dbn = DBN(variables);

[0034] 16.dbn.fit(dftrain);

[0035] (3-3) Evaluation of Dynamic Bayesian Network Model

[0036] The data organized in the training set is added to the dynamic Bayesian network model, and the windmill speed X, wind plate angle Y, fish scale screen opening Z, and loss rate Q in the training set are added to the established dynamic Bayesian model for training, and the dynamic Bayesian model predicts the Q result in the training;

[0037] Among them, p(Q1=0) represents the probability value when the loss rate in a certain row is 0, p(Q1=0)+p(Q1=1)+p(Q1=2)=1 in the 0th row, and Max(Q1) represents the state of Q when the probability of Q1 is the largest in a certain row of the table.

[0038] Preferably: the value ranges corresponding to the parameters X, Y, and Z are:

[0039] X: {0, 1, 2};

[0040] Y: {0, 1, 2};

[0041] X: {0, 1, 2}.

[0042] Preferably, the specific steps of the experimental data processing are:

[0043] Read and preprocess data; handle missing values, duplicate values, select available columns, and use the pandas library to process the entire data preprocessing process;

[0044] Standardize the data set and divide it, extract the fan speed, air distributor, fish scale screen opening and loss rate, and divide the data into two parts: training set and test set;

[0045] Data discretization processing reduces the scope of data to the state desired by the dynamic Bayesian network;

[0046] The data is processed into the data format of the data model to train the dynamic Bayesian model.

[0047] Preferably, the accuracy of the dynamic Bayesian training set is 84.43%.

[0048] Preferably, the accuracy of the dynamic Bayesian test set is 83.33%.

[0049] Technical effects and advantages of the present invention:

[0050] The present invention establishes a cleaning loss rate prediction model for a rice and wheat combine harvester based on a dynamic Bayesian network by judging the influence of cleaning operation parameters such as the angle of the wind distributor plate, the opening of the fish scale screen and the fan speed on the cleaning loss rate, as well as the influence of the cleaning loss rates at adjacent times. Through experimental data and result analysis, it can be seen that the prediction model has a high accuracy in predicting the changing trend of the cleaning loss. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a model diagram of the prediction of the cleaning loss rate based on the dynamic Bayesian network in the present invention;

[0052] Figure 2 It is a cause-and-effect diagram of the effect of the angle of the wind splitter on the cleaning loss rate in the present invention;

[0053] Figure 3 It is a cause-and-effect diagram of the effect of the fish scale screen opening on the cleaning loss rate in the present invention;

[0054] Figure 4 It is a cause-effect diagram of the influence of the fan speed on the cleaning loss rate in the present invention. DETAILED DESCRIPTION

[0055] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention are provided for the purpose of illustration and description, and are not intended to be exhaustive or to limit the present invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention, and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.

[0056] Example

[0057] In this embodiment, a prediction model for cleaning loss of a rice-wheat combine harvester based on dynamic Bayesian is provided, wherein the main cleaning operation parameters that directly affect the cleaning loss rate at the same time are the angle of the air distribution plate, the opening of the fish scale screen and the fan speed. At the same time, there is a strong correlation between the cleaning losses at adjacent times. Therefore, the following causal relationship can be defined:

[0058] Q(Q t |Q t-1 , X t , Y t , Z t )

[0059] Where: Q t is the amount of clean-up loss, the subscripts t-1 and t represent the current time and the predicted time respectively; the states of all variables in the formula are three, and the dynamic Bayesian network is expressed as expanding to different time series, such as Figure 1 As shown in FIG. 1 , it is a DBN structure diagram on two time slices.

[0060] The establishment of the cleaning loss rate model based on the dynamic Bayesian network is divided into the following steps:

[0061] (1) Select variables

[0062] Select a set of time-related variables and use them to represent a set of physical data values, for example, the variables It represents the physical data value of X in state i at time t. Then divide the time slice into time t, t-1, t-2 according to △t=1;

[0063] In this embodiment, three parameters X, Y, and Z are selected to form a variable set that affects the cleaning loss rate, where X represents the angle of the air distributor, Y represents the opening of the fish scale screen, and Z represents the fan speed. The value range of each variable must consist of a series of discrete values. Their corresponding value ranges are:

[0064] X: {0, 1, 2}

[0065] Y: {0, 1, 2}

[0066] X: {0, 1, 2}

[0067] (2) Definition of causality

[0068] (2-1) The effect of the angle of the wind splitter on the cleaning loss rate is described as a network of causal relationships between variables. represents the state i of the wind distributor angle at time t, Indicates the state i of the wind distributor angle at time t-1. There are two time slices of wind distributor angles, using the probability table The causal relationship of the wind distributor angle X on the cleaning loss rate is shown as follows: Figure 2 shown.

[0069] (2-2) The effect of fish scale screen opening on the cleaning loss rate is described as a network of causal relationships between variables. Indicates the state i of the fish scale screen opening at time t, Indicates the state i of the fish scale screen opening at time t-1. There are two time slices of fish scale screen opening, using the probability table It represents the causal relationship of the effect of fish scale screen opening Y on the cleaning loss rate, such as Figure 3 shown.

[0070] (2-3) The effect of fan speed on the cleaning loss rate is described as a network of causal relationships between variables. represents the state i of the fan speed at time t, Indicates the state i of the fan speed at time t-1. There are two time slices of fan speed, using the probability table The causal relationship between the fan speed Z and the cleaning loss rate is shown in Figure 2. Figure 4 shown.

[0071] (3) Experiment and analysis

[0072] (3-1) Experimental data

[0073] The data for this experiment comes from the field test time series data on October 27, 2018 at Nantong Farm in Haimen City, Jiangsu Province. The collected data of this sequence includes the time series data of fan speed, wind plate angle, fish scale screen opening, and cleaning impurity rate and cleaning loss rate. The data collection time interval is 1 second. 16:32:25 to 16:40:34 are selected as the training test data sequence of samples; 16:42:48 to 16:44:01 are selected as the prediction data sequence of samples. In the prediction, according to historical data, the cleaning impurity rate is continuously discretized into several states within the effective range at a step length of 0.5 for prediction (Table 1 below is part of the selected data);

[0074] Table 1- Wheat Project Data Table

[0075]

[0076] The experimental data processing steps are divided into 4 steps:

[0077] 1. Read and preprocess data. Mainly deal with missing values, duplicate values, and select available columns. During the entire data preprocessing process, we mainly use the pandas library for processing.

[0078] 2. Standardize the data set and divide it. We have a total of 205 rows of data. Extract the four columns of fan speed, air distributor, fish scale screen opening and loss rate. Divide the data into two parts: training set and test set.

[0079] 3. Data discretization processing: Narrow the range of data to the state desired by the dynamic Bayesian network.

[0080] 4. Process the data into the data format of the data model to train the dynamic Bayesian model.

[0081] (3-2) Construction of dynamic Bayesian network model

[0082] By assigning the following relationship to variables, it becomes a dynamic Bayesian network model, function DBN(), adding variables to the DBN function, and automatically generating a dynamic Bayesian model. Function dbn.fit() is used to automatically calculate the probability in dynamic Bayesian, input the training set into the dbn.fit() function, and automatically calculate the probability of X, Y, and Z in state 0, state 1, or state 2.

[0083] 1. variables =

[0084] 2.(('X', 0), ('Q', 0)),

[0085] 3.(('Y', 0), ('Q', 0)),

[0086] 4.(('Z', 0), ('Q', 0)),

[0087] 6.(('X',0),('X',1)),

[0088] 7.(('Y',0),('Y',1)),

[0089] 8.(('Z',0),('Z',1)),

[0090] 10.(('Q', 0), ('Q', 1)),

[0091] 11.(('X', 1), ('Q', 1)),

[0092] 12.(('Y', 1), ('Q', 1)),

[0093] 13.(('Z', 1), ('Q', 1))

[0094] 15.dbn=DBN(variables)

[0095] 16.dbn.fit(df_train)

[0096] (3-3) Evaluation of Dynamic Bayesian Network Model

[0097] The organized data in the training set are added to the dynamic Bayesian network model. Table 2 shows part of the selected training set;

[0098] Table 2 - Training set

[0099] (X,0) (Y,0) (Z,0) (Q,0) (X,1) (Y,1) (Z,1) (Q,1) 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 1 1 1 3 1 1 1 1 1 1 1 1 4 1 1 1 1 1 1 1 1

[0100] The windmill speed X, wind distributor Y, fish scale screen opening Z, and loss rate Q in the training set are added to the established dynamic Bayesian model for training, and the dynamic Bayesian model predicts the Q result in training. p(Q1=0) represents the probability value when the loss rate in a certain row is 0, and p(Q1=0)+p(Q1=1)+p(Q1=2)=1 in the 0th row. Max(Q1) represents the state of Q when the probability of Q1 is the largest in a certain row of the table. The details are shown in Table 3 below;

[0101] Table 3 - Probability of loss rate in different states in the training set

[0102] <![CDATA[p(Q1=0)]]> <![CDATA[p(Q1=1)]]> <![CDATA[p(Q1=2)]]> <![CDATA[Max(p(Q1))]]> 0 2.65% 84.14% 13.21% q_1 1 2.65% 84.13% 13.22% q_1 2 2.63% 84.13% 13.24% q_1 3 2.65% 84.13% 13.22% q_1 4 2.65% 84.12% 13.23% q_1

[0103] From the above experimental data, the accuracy of the dynamic Bayes training set is: 84.43%.

[0104] The present invention uses the pgmpy library in the dynamic Bayesian network to establish a dynamic Bayesian network model, puts the windmill speed, wind splitter, fish scale screen opening, and loss rate in the training set into the dynamic Bayesian network for training, obtains the learning result, and then uses the test set to test the accuracy. Through the dynamic Bayesian network method, the influencing factors are tested: the windmill speed, wind splitter, and fish scale screen opening have an impact on the prediction result of the loss rate, and the prediction results are used to study the influencing factors that affect the loss rate of the combine harvester. It is determined that the accuracy of the model test set has reached 83.33%, and the next step will continue to optimize the model accuracy.

[0105] The present invention establishes a cleaning loss rate prediction model for a rice and wheat combine harvester based on a dynamic Bayesian network by judging the influence of cleaning operation parameters such as the angle of the wind distributor plate, the opening of the fish scale screen and the fan speed on the cleaning loss rate, as well as the influence of the cleaning loss rates at adjacent times. Through experimental data and result analysis, it can be seen that the prediction model has a high accuracy in predicting the changing trend of the cleaning loss.

[0106] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field and related fields without creative work should fall within the scope of protection of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention are implemented according to the conventional means in the field unless otherwise specified and limited.

Claims

1. Based on the dynamic Bayesian rice and wheat combine harvester cleaning loss prediction model, it is characterized by: The cleaning parameters that directly affect the cleaning loss rate at the same time are the angle of the air distributor, the opening of the fish scale screen and the fan speed. There is a correlation between the cleaning losses at adjacent times, and the causal relationship is defined as follows: Q(Q t |Q t-1 ,X t ,Y t ,Z t ) Where: Q t To clear the loss amount, the subscripts t-1 and t represent the current time and the predicted time respectively; the states of all variables in the formula are three, and the dynamic Bayesian network is represented as extended to different time series.

2. The dynamic Bayesian rice-wheat combine harvester cleaning loss prediction model according to claim 1, characterized in that: The establishment of the cleaning loss prediction model is divided into the following steps: (1) Select variables Select a set of time-related variables and use them to represent a set of physical data values. Indicates the physical data value of X in state i at time t, and divides the time slice into time t, t-1, and t-2 according to Δt=1; Select parameters X, Y, and Z to form a variable set that affects the cleaning loss rate, where X represents the angle of the air distributor, Y represents the opening of the fish scale screen, and Z represents the fan speed. The value range of each variable must consist of a series of discrete values. (2) Definition of causality (2-1) The effect of the wind distributor angle on the cleaning loss rate is described as a network of causal relationships between variables: represents the state i of the wind distributor angle at time t, Indicates the state i of the wind distributor angle at time t-1. There are two time slices of wind distributor angles, using the probability table The causal relationship of the effect of the wind distributor angle X on the cleaning loss rate is shown; (2-2) The effect of fish scale screen opening on the cleaning loss rate is described as a network of causal relationships between variables. Indicates the state i of the fish scale screen opening at time t, Indicates the state i of the fish scale screen opening at time t-1. There are two time slices of fish scale screen opening, using the probability table It represents the causal relationship of the effect of fish scale screen opening Y on the cleaning loss rate; (2-3) The effect of fan speed on the cleaning loss rate is described as a network of causal relationships between variables. represents the state i of the fan speed at time t, Indicates the state i of the fan speed at time t-1. There are two time slices of fan speed, using the probability table The causal relationship of the fan speed Z on the cleaning loss rate is shown; (3) Experiment and analysis (3-1) Experimental data processing: (3-2) Construction of dynamic Bayesian network model By assigning the following relationship to variables, it becomes a dynamic Bayesian network model, function DBN(), adding variables to the DBN function to automatically generate a dynamic Bayesian model; The function dbn.fit() is used to automatically calculate the probability in dynamic Bayesian. The training set is input into the dbn.fit() function to automatically calculate the probability of X, Y, and Z being in state 0, state 1, or state 2. variables= (('X',0),('Q',0)); (('Y',0),('Q',0)); (('Z',0),('Q',0)); (('X',0),('X',1)); (('Y',0),('Y',1)); (('Z',0),('Z',1)); (('Q',0),('Q',1)); (('X',1),('Q',1)); (('Y',1),('Q',1)); (('Z',1),('Q',1)); dbn = DBN(variables); dbn.fit(df_train); (3-3) Evaluation of Dynamic Bayesian Network Model The data organized in the training set is added to the dynamic Bayesian network model, and the windmill speed X, wind plate angle Y, fish scale screen opening Z, and loss rate Q in the training set are added to the established dynamic Bayesian model for training, and the dynamic Bayesian model predicts the Q result in the training; Among them, p(Q1=0) represents the probability value when the loss rate in a certain row is 0, p(Q1=0)+p(Q1=1)+p(Q1=2)=1 in the 0th row, and Max(Q1) represents the state of Q when the probability of Q1 is the largest in a certain row of the table.

3. The dynamic Bayesian rice-wheat combine harvester cleaning loss prediction model according to claim 2, characterized in that: The value ranges corresponding to the parameters X, Y, and Z are: X:{0,1,2}; Y:{0,1,2}; X:{0,1,2}。 4. The dynamic Bayesian rice-wheat combine harvester cleaning loss prediction model according to claim 2, characterized in that: The specific steps of the experimental data processing are: Read and preprocess data; handle missing values, duplicate values, select available columns, and use the pandas library to process the entire data preprocessing process; Standardize the data set and divide it, extract the fan speed, air distributor, fish scale screen opening and loss rate, and divide the data into two parts: training set and test set; Data discretization processing reduces the scope of data to the state desired by the dynamic Bayesian network; The data is processed into the data format of the data model to train the dynamic Bayesian model.

5. The dynamic Bayesian rice-wheat combine harvester cleaning loss prediction model according to any one of claims 1-4, characterized in that: The accuracy of the dynamic Bayesian training set is: 84.43%.

6. The dynamic Bayesian rice-wheat combine harvester cleaning loss prediction model according to any one of claims 1-4, characterized in that: The accuracy of the dynamic Bayesian test set is: 83.33%.