Agricultural material dynamic regulation and control cleaning method and system based on accumulation condition prediction

Abstract

The invention discloses an agricultural material dynamic regulation and control cleaning method and system based on accumulation condition prediction. The method comprises the steps that one-dimensional air volume array data generated by all first air volume sensors and two-dimensional air volume array data generated by all second air volume sensors are obtained; calculating a comprehensive non-uniformity index of the air volume below the screen surface based on the one-dimensional air volume array data, judging whether the comprehensive non-uniformity index exceeds a first preset threshold value or not, and if so, compensating data in the two-dimensional air volume array data based on the comprehensive non-uniformity index; according to the two-dimensional air volume array data, calculating a feature vector reflecting the unbalance of the materials on the screen surface; and based on the feature vector and a pre-trained prediction model, predicting the stacking height of the material to obtain a predicted stacking height, and adjusting the rotating speed of the material scattering roller. The material accumulation condition on the screen surface can be effectively predicted based on the air volume data on the upper side and the lower side of the cleaning screen, the materials are properly scattered accordingly, and the material accumulation problem in the traditional cleaning process is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control of material sorting, and in particular to a method and system for dynamically regulating and sorting agricultural materials based on prediction of accumulation conditions. Background Art

[0002] Agricultural material cleaning is a core link in agricultural processing, which directly affects the purity and storage quality of grains. Screening equipment separates impurities through airflow, but local accumulation on the screen surface will destroy the uniformity of air volume distribution: airflow obstruction in the accumulation area leads to impurities residue, and excessive air volume in adjacent areas causes excessive grain scattering or throwing losses. Traditional methods rely on fixed air volume or manual adjustment, and cannot monitor dynamic accumulation in real time, resulting in low cleaning efficiency and limited effect.

[0003] In the prior art, the applicant's prior application CN 115176603 A discloses a material dispersion and diversion mechanism for the cleaning screen surface used on the screen surface of the cleaning screen. Although the material on the screen surface can be dispersed evenly, the start and stop and the rotation speed need to be manually controlled based on experience, and the control is not intelligent enough.

[0004] Patent CN 208912557 U provides a wind-screen type grain cleaning test device, which is equipped with arrayed wind speed sensors on both sides of the vibrating screen, which monitor the wind speed distribution near the fan outlet and the screen surface in real time, analyze the data through a programmable controller and dynamically adjust the cleaning parameters to ensure that the airflow evenly covers the screen surface. This technology lacks multi-dimensional data fusion, dynamic compensation and intelligent prediction capabilities, resulting in insufficient adaptability and robustness of the cleaning process, and cannot effectively solve the problem of material accumulation under complex working conditions. Summary of the invention

[0005] Purpose of the invention: In order to overcome the deficiencies in the prior art, the present invention provides a method and system for dynamically controlling the cleaning of agricultural materials based on the prediction of the accumulation condition, which can effectively predict the material accumulation condition on the screen surface based on sensor data and break up the material as needed to improve the cleaning effect and efficiency.

[0006] Technical solution: To achieve the above purpose, the method for dynamically regulating and cleaning agricultural materials based on the prediction of the accumulation situation of the present invention comprises:

[0007] Get the following below the cleaning screen r One-dimensional wind volume array data generated by the first wind volume sensor , For the k The data generated by the first air volume sensor; and obtaining the data above the cleaning screen p OK q List all two-dimensional wind volume array data generated by the second wind volume sensor , The data generated by the second air volume sensor in the i th row and j th column;

[0008] Based on the one-dimensional air volume array data, calculate the comprehensive unevenness index Γ of the air volume distribution below the sieve surface of the cleaning sieve;

[0009] Judge whether Γ exceeds the first preset threshold Γ0. If so, compensate the data in the two-dimensional air volume array data based on the comprehensive unevenness index, otherwise maintain the data in the two-dimensional air volume array data unchanged;

[0010] According to the two-dimensional air volume array data, calculate the eigenvector reflecting the unevenness of the sieve surface material , where: is the global standard deviation, is the local maximum deviation, is the maximum variance of the sub-region;

[0011] Based on the eigenvector and the pre-trained prediction model, predict the stacking height of the material to obtain the predicted stacking height , and adjust the rotation speed of the material dispersion drum through the following control strategy n :

[0012] ;

[0013] where: is the proportional coefficient, is the stacking height threshold.

[0014] Furthermore, each parameter in the eigenvector is calculated based on the following formula:

[0015] Global standard deviation ; where: average wind speed ;

[0016] Local maximum deviation ;

[0017] Sub-region maximum variance ;

[0018] Among them, , divide the two-dimensional air volume array data into a × b sub-regions, is the number of sensors in the M th sub-region, is the​M The average air volume within a sub-region.

[0019] Further, the comprehensive non-uniformity index Γ is calculated based on the following formula:

[0020]

[0021] Wherein, is the average air volume.

[0022] Further, compensating the data in the two-dimensional air volume array data based on the comprehensive non-uniformity index includes:

[0023] Calculating the normalized non-uniformity weight at each position below the sieve surface of the cleaning sieve ;

[0024] Compensating the two-dimensional air volume array data of the second air volume sensor specifically: assigning the value obtained from the formula ( ) to in for data update; wherein, is the compensation ratio coefficient, is the normalized non-uniformity weight corresponding to the i-th row, is the air volume compensation amplitude.

[0025] Further, the prediction model is created based on the following method:

[0026] Obtaining experimental data , where is the actual stacking height of the material on the sieve , is the feature vector, N is the total number of groups of experimental data;

[0027] Based on the experimental data and the SVR surrogate model for data training to obtain the prediction model; the objective function of the SVR surrogate model is:

[0028] ;

[0029] Wherein, is the weight vector, is the bias term, is the penalty factor, and are slack variables; is the total number of trials;

[0030] Adopting a radial basis function kernel ; wherein is the kernel width parameter.

[0031] Furthermore, the sub-area division rules include:

[0032] Determine the size of the subregion c × d ,in c , d are the numbers of the second air volume sensors included in the horizontal and vertical directions respectively;

[0033] Determine the number of overlapping rows and columns when selecting sub-regions e and f ;

[0034] Based on the sub-region size and the number of overlapping rows and columns, all combinations that meet the conditions in all the second air volume sensors are traversed to form the sub-region that meets the conditions.

[0035] The agricultural material dynamic control and cleaning system based on the prediction of accumulation situation includes:

[0036] Cleaning screens, used for cleaning agricultural materials to separate seeds from foreign matter;

[0037] A one-dimensional linear array layout located below the cleaning screen r a first air volume sensor;

[0038] The press located above the cleaning screen p × q A plurality of second air volume sensors arranged in a square array;

[0039] Material breaking drum is used to break up the accumulated materials on the cleaning screen;

[0040] A controller is connected to the first air volume sensor, the second air volume sensor and the material breaking drum, and is capable of implementing the above-mentioned agricultural material dynamic control and cleaning method based on accumulation situation prediction.

[0041] Beneficial effects: The method and system for dynamically regulating and cleaning agricultural materials based on the prediction of the accumulation situation of the present invention have the following beneficial effects:

[0042] (1) By acquiring the air volume array data above and below the cleaning screen in real time and combining it with the comprehensive unevenness index to dynamically compensate for the two-dimensional air volume distribution, a multidimensional feature vector is constructed using the global standard deviation, local maximum deviation and sub-region maximum variance. The material accumulation height is predicted based on the proxy model, and the drum speed is adaptively adjusted. This effectively solves the material accumulation problem caused by uneven air volume distribution in the traditional cleaning process, improves the cleaning efficiency and quality, and reduces the need for manual intervention through data-driven closed-loop control, thereby enhancing the system's dynamic response capability and stability to complex working conditions.

[0043] (2) The three parameters in the characteristic vector obtained based on the two-dimensional air volume array data can effectively reflect the distribution of air volume on the screen surface in the three dimensions of overall, local, and extreme differences. They are highly representative and can effectively reflect the distribution and accumulation of materials.

[0044] (3) The comprehensive non-uniformity index Γ comprehensively evaluates the overall discreteness and local mutation intensity of the air volume distribution below the screen surface by combining the global standard deviation and the absolute mean of the air volume difference between adjacent sensors. It can more scientifically reflect the true uniformity of the air flow distribution and avoid one-sidedness, thereby accurately triggering the dynamic compensation of the second sensor data, effectively correcting the prediction error caused by uneven air flow under the screen, and significantly improving the robustness of the stacking height prediction and the adaptability of the cleaning control.

[0045] (4) By dynamically adjusting the compensation parameters of the second air volume sensor based on the air volume conditions at different positions below the screen surface, the data of the second air volume sensor is compensated based on the data of the first air volume sensor, thereby improving the accuracy of subsequent stacking height prediction.

[0046] (5) When dividing sub-areas, the number of overlapping rows and columns is set to ensure data continuity between adjacent sub-areas, avoid edge effects caused by division, enhance the ability to capture local airflow mutations and abnormal accumulation, improve the comprehensiveness of feature extraction and model prediction accuracy, and adapt to different material distribution patterns to optimize the refinement level of cleaning and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a structural diagram of a dynamic control and cleaning system for agricultural materials based on the prediction of accumulation conditions;

[0048] Figure 2 is a layout diagram of the second wind volume sensor in the top view;

[0049] Figure 3 The figure is a flow chart of the method for dynamically controlling and cleaning agricultural materials based on the prediction of the accumulation situation.

[0050] In the figure: 1-cleaning screen; 2-first air volume sensor; 3-second air volume sensor; 4-material breaking drum; 5-fan. DETAILED DESCRIPTION

[0051] The present invention will be further described below in conjunction with the accompanying drawings.

[0052] like Figure 1The agricultural material dynamic control cleaning system based on the prediction of the accumulation situation shown in the figure comprises a cleaning screen 1 for cleaning agricultural materials to separate seeds from impurities, r first air volume sensors 2 arranged in a one-dimensional linear array below the cleaning screen 1, and a plurality of second air volume sensors 3 arranged in a p×q square array above the cleaning screen 1, as shown in FIG. Figure 2 As shown; in addition, a material breaking drum 4 is installed above the cleaning screen for breaking up the aggregated materials on the cleaning screen 1; the cleaning system also includes a fan 5 for providing a cleaning airflow.

[0053] The system also includes a controller, which is connected to the first air volume sensor 2, the second air volume sensor 3 and the material scattering drum 4, and can implement the following agricultural material dynamic control and cleaning method based on the prediction of the accumulation situation.

[0054] As Figure 3 The agricultural material dynamic control and sorting method based on accumulation situation prediction shown includes the following steps S101-S105:

[0055] Step S101, obtain the r One-dimensional wind volume array data generated by the first wind volume sensor 2 , For the k The data generated by the first air volume sensor 2; and obtaining the data above the cleaning screen 1 p Line q List all two-dimensional wind volume array data generated by the second wind volume sensor 3 , For the i Line j The data generated by the second wind volume sensor 3 of the column;

[0056] Step S102: based on the one-dimensional wind volume array data Calculating the comprehensive unevenness index Γ of the air volume distribution below the screen surface of the cleaning screen 1;

[0057] Step S103, determining whether Γ exceeds a first preset threshold Γ0, if yes, then performing an analysis on the two-dimensional air volume array data based on the comprehensive non-uniformity index. The data in the data are compensated, otherwise the two-dimensional air volume array data is maintained The data in remains unchanged;

[0058] Step S104: according to the two-dimensional wind volume array data Calculate the characteristic vector that reflects the imbalance of the material on the screen surface , where: is the global standard deviation, is the local maximum deviation, is the maximum variance of the sub-region;

[0059] Step S105: Based on the feature vector and a pre-trained prediction model, predict the stacking height of the material to obtain the predicted stacking height , and adjust the rotation speed of the material dispersion roller 4 through the following control strategy:

[0060] ;

[0061] where: is the proportionality coefficient, is the stacking height threshold.

[0062] This method dynamically compensates the two-dimensional air volume distribution by combining the comprehensive inhomogeneity index through real-time acquisition of the air volume array data above and below the cleaning sieve 1, constructs a multi-dimensional feature vector using the global standard deviation, local maximum deviation, and the maximum variance of the sub-region, and predicts the stacking height of the material based on the surrogate model to achieve the adaptive adjustment of the roller rotation speed, effectively solving the problem of material stacking caused by uneven air volume distribution in the traditional cleaning process, improving the cleaning efficiency and quality. At the same time, the data-driven closed-loop control reduces the need for manual intervention, enhancing the dynamic response ability and stability of the system under complex working conditions.

[0063] Preferably, the parameters in the feature vector in the above step S104 are calculated based on the following formulas:

[0064] Global standard deviation ; where: average wind speed ;

[0065] Local maximum deviation ;

[0066] Maximum variance of the sub-region ;

[0067] where, , divide the two-dimensional air volume array data into a × b sub-regions, is the number of sensors in the M th sub-region, is the average air volume in the M th sub-region, is the data generated by the second air volume sensor 3 in the m rd row and n th column in the sub-region.

[0068] Among the above three parameters, the global standard deviation reflects the overall dispersion degree of the air volume distribution on the entire sieve surface and can indicate the uniformity of the overall air volume. However, when materials are locally piled up on the sieve surface, the air volume in some sub-regions may deviate significantly from the average value. Therefore, by dividing the sieve surface into sub-regions and calculating the variance of each sub-region, and selecting the maximum value among them, the maximum variance of the sub-region can analyze the uniformity of the local air volume distribution in detail. It highlights the region with the most uneven air volume distribution, that is, the local region where material piling may be more serious. Even if the global standard deviation is low, the maximum variance of the sub-region can detect these local anomalies. The local maximum deviation plays a role in quickly and efficiently capturing extreme differences in air volume distribution throughout the method, providing important feature support for solving the problem of material piling.

[0069] It can be seen that the three parameters in the eigenvector obtained from the two-dimensional air volume array data can effectively reflect the air volume distribution on the sieve surface in three dimensions: overall, local, and extreme differences. They have strong representativeness and can effectively reflect the distribution and piling status of materials.

[0070] Preferably, the comprehensive non-uniformity index Γ in the above step S102 is calculated based on the following formula:

[0071]

[0072] Where is the average air volume. The first term on the right side of the above formula is the global standard deviation, and the second term is the absolute average value of the air volume difference between adjacent sensors.

[0073] By combining the global standard deviation and the absolute average value of the air volume difference between adjacent sensors, the comprehensive non-uniformity index Γ comprehensively evaluates the overall dispersion degree and local mutation intensity of the air volume distribution below the sieve surface, can more scientifically reflect the true uniformity of the air flow distribution, avoids one-sidedness, thus accurately triggers the dynamic compensation for the data of the second sensor, effectively corrects the prediction error caused by uneven air flow under the sieve, and significantly improves the robustness of the piling height prediction and the adaptability of the cleaning regulation.

[0074] Preferably, the compensation for the two-dimensional air volume array data based on the comprehensive non-uniformity index in the above step S103 includes the following steps S201 - S202:

[0075] Step S201, calculate the normalized non-uniformity weight at each position below the sieve surface of the cleaning sieve 1 ;

[0076] Step S202, compensate the two-dimensional air volume array data of the second air volume sensor 3 , specifically: assign the value obtained from the formula ( ) to in perform data update; among them, is the compensation ratio coefficient, is the i corresponding normalized non-uniformity weight of the wind volume compensation amplitude.

[0077] In step S202, use to represent the value obtained from the arithmetic expression ( ). During the compensation process, constrain the change range of to ensure that the compensated wind volume data will not introduce new non-uniformity, that is:

[0078] ;

[0079] Among them, and are respectively the reasonable upper and lower limits of the wind volume data of the rd second wind volume sensor 3.

[0080] Through the above method of dynamically adjusting the compensation parameters of the second wind volume sensor 3 based on the wind volume conditions at different positions below the sieve surface, compensating the data of the second wind volume sensor 3 based on the data of the first wind volume sensor 2 can improve the accuracy of subsequent stacking height prediction.

[0081] Preferably, the prediction model in the above step S105 is created based on the following method:

[0082] Step S301, obtain experimental data , where is the actual stacking height of the material on the sieve surface , is the feature vector, N is the total number of groups of experimental data; the feature vector here is calculated using the algorithms in the above steps S101 - S104;

[0083] Step S302, perform data training on the experimental data and the SVR surrogate model to obtain the prediction model; the objective function of the SVR surrogate model is:

[0084] ;

[0085] Among them, is the weight vector, is the bias term, is the penalty factor, and are the slack variables, , ; is the total number of tests; here, all parameters are local variables;

[0086] and satisfy the following constraint conditions:

[0087] For all , , and, ; where is an insensitive parameter that defines the tolerance range of the prediction error; is a mathematical tool in machine learning that non-linearly maps data to a high-dimensional space.

[0088] Adopt a radial basis function kernel ; where is the kernel width parameter.

[0089] This technology effectively balances the model complexity and training error through the SVR surrogate model combined with the regularization term and the relaxation variable penalty mechanism, maps non-linear features to a high-dimensional space using the radial basis kernel function, significantly improves the generalization ability and accuracy of the stacking height prediction; based on the experimental data training to ensure that the model fits the actual cleaning conditions, adapts to the changing material distribution and air flow interference, provides a reliable decision-making basis for dynamic adjustment, and overall enhances the stability and intelligent level of the cleaning system.

[0090] Preferably, the division rule of the sub-regions includes the following steps S401 - S402:

[0091] Step S401, determine the size of the sub-region c × d , where c , d are the numbers of the second air volume sensors 3 included in the horizontal and vertical directions respectively;

[0092] Step S402, determine the number of overlapping rows and columns when selecting a sub-region e and f ; that is, when selecting a sub-region, there are e rows of sensors that overlap between two adjacent sub-regions in the vertical direction, and there are f columns of sensors that overlap between two adjacent sub-regions in the horizontal direction. e Less than d , f Less than c .

[0093] Step S403, based on the sub-region size and the number of overlapping rows and columns, traverse all combinations that meet the conditions among all the second air volume sensors 3, and form the sub-regions that meet the conditions.

[0094] Through the above method, by setting the number of overlapping rows and columns, the data continuity between adjacent sub-regions is ensured, the edge effect caused by division is avoided, the ability to capture local airflow mutations and abnormal accumulations is enhanced, the comprehensiveness of feature extraction and the model prediction accuracy are improved, while adapting to different material distribution patterns and optimizing the refinement level of cleaning control.

[0095] The above are only the preferred embodiments of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. An agricultural material dynamic regulation and cleaning method based on prediction of stacking conditions, characterized in that, The method comprises: Obtain the data of the one-dimensional air volume array generated by r the first air volume sensors below the cleaning sieve , which is the data generated by the k th first air volume sensor; and obtain the two-dimensional air volume array data generated by all the second air volume sensors in p rows q and columns above the cleaning sieve , i which is the data generated by the second air volume sensor in the j th row and th column; Based on the one-dimensional air volume array data Calculate the comprehensive unevenness index Γ of the air volume distribution below the sieve surface of the cleaning sieve; Determine whether Γ exceeds the first preset threshold Γ0. If so, compensate the data in the two-dimensional air volume array data based on the comprehensive non-uniformity index, otherwise maintain the data in the two-dimensional air volume array data unchanged; Otherwise, keep the data in the two-dimensional air volume array data unchanged; According to the two-dimensional air volume array data Calculate the eigenvector reflecting the imbalance of materials on the reaction sieve surface , where: is the global standard deviation, is the local maximum deviation, is the maximum variance of the sub-region; Based on the feature vector and a pre-trained prediction model, predict the stacking height of the material to obtain the predicted stacking height , and adjust the rotation speed of the material dispersion roller through the following control strategy n : ； Wherein: is a proportionality coefficient, is a stacking height threshold value.

2. The dynamic regulation and cleaning method for agricultural materials based on prediction of stacking conditions according to claim 1, characterized in that Each parameter in the feature vector is calculated based on the following formula: Global standard deviation ; wherein: average wind speed ; Local maximum deviation ; Maximum variance of sub-region ; Among them, the two-dimensional air volume array data is divided into a × b sub-regions, is the number of sensors in the M th sub-region, is the average air volume in the M th sub-region.

3. The agricultural material dynamic regulation and cleaning method based on prediction of stacking situation according to claim 1, wherein The comprehensive inhomogeneity index Γ is calculated based on the following formula: 。 4. Among them, is the average air volume.

5. The dynamic regulation and cleaning method for agricultural materials based on prediction of stacking conditions according to claim 3, characterized in that, Compensating the data in the two-dimensional air volume array data based on the comprehensive inhomogeneity index, including: ​ Calculate the normalized non-uniformity weight at each position below the sieve surface of the cleaning sieve ; Compensate the two-dimensional air volume array data of the second air volume sensor , specifically: assign the value obtained from the formula ( ) to in for data update; where is the compensation ratio coefficient, is the normalized non-uniformity weight corresponding to the i th row, and is the air volume compensation amplitude.

6. The dynamic regulation and cleaning method of agricultural materials based on prediction of stacking conditions according to claim 1, characterized in that The prediction model is created based on the following method: Obtain experimental data , where is the actual stacking height of the material on the sieve surface , is the feature vector, N is the total number of groups of experimental data; The prediction model is obtained by performing data training based on the experimental data and the SVR proxy model; the objective function of the SVR proxy model is: ； Among them, is the weight vector, is the bias term, is the penalty factor, and are slack variables; is the total number of trials; Using a radial basis function kernel ; where is the kernel width parameter.

7. The method for dynamically regulating and cleaning agricultural materials based on prediction of stacking conditions according to claim 2, wherein The sub-area division rules include: Determine the size of the sub-region c × d , where c 、 d are respectively the numbers of the second air volume sensors included in the horizontal and vertical directions; Determine the number of overlapping rows and columns when a selected sub-region is determined e and f ; Based on the sub-region size and the number of overlapping rows and columns, all combinations that meet the conditions in all the second air volume sensors are traversed to form the sub-region that meets the conditions.

8. An agricultural material dynamic regulation and cleaning system based on prediction of stacking conditions, characterized in that, It includes: Cleaning screens, used for cleaning agricultural materials to separate seeds from foreign matter; One-dimensional linear array layout located below the cleaning sieve r first air volume sensors; Above the cleaning sieve, according to p × q A plurality of second air volume sensors arranged in a square array; Material breaking drum is used to break up the accumulated materials on the cleaning screen; A controller is connected to the first air volume sensor, the second air volume sensor and the material breaking drum, and is capable of implementing the agricultural material dynamic control and cleaning method based on stacking condition prediction as described in any one of claims 1 to 6.

Images