A double variable fertilizer machine fertilization decision method

Abstract

This invention relates to a bivariate fertilizer application decision-making method for a fertilizer applicator, comprising: data preparation, fertilizer application prediction model training, and fertilizer application decision solving. Specifically, data preparation mainly includes initial data collection, outlier handling based on the Grubbs criterion, and data segmentation; fertilizer application prediction model training mainly includes dataset partitioning, data normalization, and model training, wherein the model is trained using a multi-core extreme learning machine, and the activation function weight vector of the multi-core extreme learning machine is optimized using a compression factor particle swarm optimization algorithm; fertilizer application decision solving mainly includes establishing a fertilizer performance model and solving the model; wherein a problem model is established with fertilizer application accuracy, fertilizer application uniformity, equipment adjustment time, and fertilizer breakage rate as objectives, and an improved NSGA-III is used to solve the problem model to obtain a set of optimal fertilizer application decisions corresponding to the target fertilizer application amount.

Description

Technical Field

[0001] This invention relates to a bivariate fertilizer applicator fertilization decision-making method, belonging to the field of intelligent agriculture technology; and particularly to an optimized fertilizer applicator fertilization decision-making method. Background Technology

[0002] As an important component of precision agriculture, variable fertilization can provide optimal fertilizer input to the areas of the field that require fertilization, which helps increase crop yields, reduce environmental pollution, lower agricultural costs, and increase economic income.

[0003] Currently, mainstream solid particle variable discharge fertilizer applicators can be divided into single-variable and dual-variable discharge fertilizer applicators based on the number of controlled variables. While single-variable discharge fertilizer applicators are easier to control than dual-variable discharge fertilizer applicators, they exhibit pulsating fertilization at low speeds, resulting in poor fertilization uniformity. Although dual-variable discharge fertilizer applicators can solve this problem to some extent, the single fertilizer application rate may correspond to multiple combinations of different discharge port openings and discharge shaft speeds. If non-optimal speeds and openings are used during fertilization, the fertilization performance will also deteriorate. Summary of the Invention

[0004] The purpose of this invention is to provide a fertilization decision-making method for a bivariate fertilizer applicator, which solves the problem that existing strategies are unable to obtain the optimal combination of feed inlet opening and fertilizer discharge shaft speed.

[0005] To achieve the above objectives, the present invention includes:

[0006] The technical solution of a bivariate fertilizer application decision-making method for a fertilizer applicator according to the present invention includes the following steps:

[0007] 1) Prepare training data; the training data includes several sets of fertilizer discharge shaft speed and discharge port opening of the fertilizer applicator as model inputs, and the corresponding fertilizer application amount of the fertilizer applicator as model outputs;

[0008] 2) The improved extreme learning machine model is trained using training data to obtain a fertilizer application prediction model;

[0009] 3) Establish a fertilization performance model with at least one of the following as objectives: highest fertilization accuracy, highest fertilization uniformity, highest equipment adjustment timeliness, and lowest fertilizer breakage rate. Solve the fertilization performance model with different combinations of feed port opening and fertilizer discharge shaft rotation speed as constraints to obtain the optimal solution set of rotation speed and opening combinations corresponding to the target fertilizer application rate; the corresponding target fertilizer application rate is obtained based on the fertilizer application rate prediction model.

[0010] 4) During fertilization, select the combination of rotation speed and opening degree that satisfies the fertilizer application requirements from the optimal solution set as the fertilizer application decision control fertilizer applicator.

[0011] This invention pre-calculates the optimal fertilization decision for each region based on the target fertilization rate in the fertilization prescription diagram, thereby simplifying the control process while improving the fertilization performance of the bivariate fertilizer applicator.

[0012] Studies have shown that increasing the rotational speed of the fertilizer discharge shaft leads to an excessively high fertilizer breakage rate. Excessive breakage results in fertilizer waste and increased economic costs. Furthermore, it prevents timely discharge of broken material, causing it to accumulate in the gap between the bottom of the fertilizer discharger and the discharge wheel, reducing the discharger's volume and affecting fertilization accuracy. This invention addresses this issue by using fertilizer breakage rate as a target for optimizing fertilization decisions. Combined with existing research's focus on optimizing fertilization accuracy, uniformity, and equipment adjustment time, this invention effectively reduces fertilizer breakage rate and thereby increases fertilization accuracy.

[0013] This invention optimizes fertilization decisions by targeting fertilization accuracy, fertilization uniformity, equipment adjustment time, and fertilizer breakage rate, and obtains the optimal combination of rotation speed and opening that can simultaneously meet the above conditions.

[0014] Furthermore, the preparation of the training data includes: collecting initial data on fertilizer application amount using a fertilizer applicator frame with different combinations of rotation speed and opening, then performing outlier detection and removal on the initial data, and segmenting the data after removing outliers according to the degree of influence of the feed opening on the fertilizer growth trend; each segment of data is used as the training set, validation set, and test set for training the extreme learning machine model.

[0015] Considering that the amount of fertilizer applied by the fertilizer applicator is highly sensitive to changes in the opening when the applicator is operating at a small opening, and the growth rate of fertilizer application varies drastically, this invention segments the initial data according to the degree of influence of the feed opening opening on the fertilizer growth trend, thus avoiding the impact of differences in fertilizer growth rate on the accuracy of the fertilizer rate prediction model.

[0016] Furthermore, outlier detection of the initial data is performed using the Grubbs criterion, and the outliers are filled in by the mean of the data sample after removal.

[0017] Furthermore, the data is segmented as follows: First, the average growth rate of the fertilization rate corresponding to each feed inlet opening degree and its previous opening degree is calculated, specifically expressed as:

[0018]

[0019] In the formula, For the first The average fertilization growth rate corresponding to each opening degree compared to the previous opening degree. For the first The fertilization rate corresponding to each opening degree and the j-th rotation speed This represents the fertilization rate corresponding to the previous opening degree and the j-th rotation speed. The total number of rotational speeds;

[0020] This leads to the average growth rate of fertilization rate. The average growth rate of fertilization rate Group the data and find the maximum and minimum variances of all possible groups. Use the median of the maximum and minimum variances as the segment reference values. Take the data group whose variance differs the least from the segment reference value as one segment and the remaining data as another segment.

[0021] Because online monitoring technology for variable fertilization is still immature, current research on fertilization decision optimization mostly employs machine learning to train fertilization rate prediction models using raw data obtained from calibration experiments, in order to predict unknown fertilization amounts. While some researchers have achieved certain results in fertilization rate prediction models, there are still shortcomings in the preprocessing of raw data. Since raw data can significantly influence the degree to which the fertilization rate prediction model matches the actual fertilization process, this invention considers data preprocessing as one of the main steps in fertilization decision optimization.

[0022] Furthermore, training the improved extreme learning machine model yields a fertilizer application prediction model including:

[0023] The construction of a multi-core single-hidden-layer extreme learning machine model is specifically represented as follows:

[0024]

[0025] in, The connection weights between the hidden layer and the output layer. For the first One activation function, The weights of the activation function, , These are the connection weights between neurons in the input layer and the hidden layer. The threshold of the hidden layer neurons. For the input sample, This represents the number of neurons in the hidden layer. The number of activation functions. This is the output sample.

[0026] In training the fertilizer application prediction model, this invention divides each segment of preprocessed training data into training, validation, and test sets. To avoid disrupting the original data distribution during dataset partitioning, each segment is first divided into several small intervals, and one sample is extracted from each interval as the test set data. The validation set data is partitioned using the same method. Furthermore, the preprocessed training data is normalized to scale the fertilizer discharge shaft speed, discharge port opening, and their corresponding fertilizer application rates to between 0 and 1, preventing the different units of measurement from affecting the model's prediction accuracy during training. Data normalization can be performed first, followed by partitioning the normalized data into training, validation, and test sets; alternatively, the dataset can be partitioned first, followed by normalization of each dataset. The specific order is not limited in this invention.

[0027] Furthermore, an adaptive compression factor particle swarm optimization algorithm is employed to minimize the mean squared error of the model on the validation set, adjusting the activation function weights. Optimization is performed to obtain the optimal activation function weight vector.

[0028] Furthermore, the velocity update formula for the adaptive compression factor particle swarm is:

[0029]

[0030] Among them, compression factor , , and For learning factors.

[0031] Therefore, it is necessary to select an optimal set of activation function weight vectors. This invention uses an adaptive compression factor particle swarm optimization algorithm to optimize the weight vectors with the goal of minimizing the mean square error of the model on the validation set. This ensures that the activation function weight vectors are optimal, improves the weight differences of different activation functions in the multi-core extreme learning machine, and improves the prediction performance of the final trained model.

[0032] Furthermore, the adaptive compression factor particle swarm algorithm dynamically adjusts the learning factor according to the change in the number of iterations. and and through , For compressibility factor Make dynamic adjustments. , The updated formula is:

[0033]

[0034]

[0035] in, , This represents the current iteration number. To minimize the number of iterations, This represents the maximum number of iterations.

[0036] Furthermore, among those with the highest fertilization accuracy, highest fertilization uniformity, highest equipment adjustment timeliness, and lowest fertilizer breakage rate:

[0037] The objective of achieving the highest fertilization accuracy is defined as minimizing the absolute difference between the predicted fertilization amount from the fertilization amount prediction model and the actual fertilization amount required by the fertilization prescription map, specifically expressed as:

[0038]

[0039] in, The rotational speed of the fertilizer discharge shaft. For the opening of the feed inlet, The required amount of fertilizer, This is the predicted value for fertilizer application. This represents the maximum permissible error for the fertilizer applicator.

[0040] The optimization target model for fertilizer uniformity, equipment adjustment timeliness and fertilizer breakage rate is defined by converting the rotation speed and opening degree from rectangular coordinates to polar coordinates. In the rectangular coordinate system, the horizontal axis is the opening degree and the vertical axis is the rotation speed.

[0041] The target for achieving the highest fertilization uniformity is defined as having the largest polar angle in polar coordinates, specifically expressed as:

[0042] =

[0043] in, This represents the polar angle of the rotational speed and opening degree in polar coordinates;

[0044] The objective for maximizing the timeliness of equipment adjustment is defined as having the minimum absolute value of the difference between the polar angles corresponding to two fertilization rates and rotation speeds in polar coordinates. Specifically, it is expressed as follows:

[0045] =

[0046] in, This represents the polar angle of the rotation speed opening corresponding to the current fertilization rate in polar coordinates. This represents the polar angle of the rotation speed opening corresponding to the previous fertilization rate in polar coordinates.

[0047] The target for minimizing fertilizer breakage rate is defined as minimizing the absolute value of the difference between the polar angle of the current fertilization rate corresponding to the rotation speed and opening degree in polar coordinates and the polar angle of the rotation speed of 100 r / min and the opening degree corresponding to the previous fertilization rate in polar coordinates. Specifically, it is expressed as follows:

[0048] =

[0049] in, The polar angle in polar coordinates is the value of the rotation speed of 100 r / min and the opening degree corresponding to the previous fertilization opening degree.

[0050] Furthermore, the fertilization performance model is solved using the improved NSGA-III; the improved NSGA-III improves the reference point association operation by selecting individuals with the shortest vertical distance to the reference line and the shortest distance from the intersection of the reference line and the reference point to the reference point for association.

[0051] This invention establishes and solves a fertilization performance model for optimizing fertilization performance. Specifically, it establishes a multi-problem optimization model with objectives of fertilization accuracy, fertilization uniformity, equipment adjustment time, and fertilizer breakage rate. The models for fertilization uniformity, equipment adjustment time, and fertilizer breakage rate are established by performing polar coordinate transformation on rotation speed and opening degree, and establishing the corresponding optimization model based on the polar angle in polar coordinates. Finally, the problem model is solved based on the improved NSGA-III multi-objective evolutionary algorithm with the feed inlet opening degree and fertilizer discharge shaft rotation speed as constraints, and the optimal solution set for the corresponding target fertilization amount is obtained.

[0052] Considering the complexity of the control process of a dual-variable fertilizer applicator and the limited computing power of the on-board fertilizer applicator controller, this invention provides a fertilization decision-making method for a dual-variable fertilizer applicator. This method pre-determines the corresponding rotation speed and opening degree based on the target fertilizer amount required for each area in the fertilizer prescription diagram. This allows the applicator to simultaneously meet four fertilization performance indicators: fertilization accuracy, fertilization uniformity, equipment adjustment time, and fertilizer breakage rate. This significantly simplifies the control process of the dual-variable fertilizer applicator, reduces the computing burden on the on-board controller, and enables the fertilizer applicator to complete fertilization operations quickly and efficiently. Attached Figure Description

[0053] Figure 1 A diagram illustrating the overall implementation steps of this invention;

[0054] Figure 2 A data preprocessing flowchart provided for embodiments of the present invention;

[0055] Figure 3 This is a flowchart illustrating the training process of the fertilizer application prediction model provided in an embodiment of the present invention.

[0056] Figure 4This is a flowchart illustrating the fertilization decision optimization process provided in an embodiment of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0058] Figure 1 The following is a flowchart of the overall steps of a bivariate fertilizer applicator fertilization decision-making method provided by the present invention: First, initial indoor data is collected through the fertilizer applicator platform. The Grubbs criterion is used to perform outlier detection on the collected data. The average growth rate of fertilizer application corresponding to each opening degree is calculated. The data is then segmented according to the difference in the average growth rate.

[0059] The improved Extreme Learning Machine was used to train the data of different groups after segmentation, and a fertilizer application prediction model was obtained.

[0060] An optimization problem model was established with the objectives of fertilization accuracy, fertilization uniformity, equipment adjustment timeliness, and fertilizer breakage rate. The improved NSGA-III algorithm was used to solve the problem model and obtain the final fertilization decision.

[0061] 1. Data preparation.

[0062] More specifically, the data preprocessing flowchart of a bivariate fertilizer application decision-making method in this embodiment is as follows: Figure 2 As shown, the initial data of fertilizer application were first collected by using the fertilizer applicator frame with different combinations of rotation speed and opening. The rotation speed was adjusted from 10 to 150 r / min and the opening was adjusted from 6 to 28 mm. A total of 345 sets of data were collected, with each set containing 20 samples.

[0063] After data collection, outlier detection was performed using the Grubbs criterion. Since the Grubbs criterion requires the data to follow a normal distribution, the Shapiro-Wilks test was first used to verify the normality of the data at a significance level of 0.05 before outlier detection. If the test result p... If the value is 0.05, then the data set is considered to follow a normal distribution. For data that follows a normal distribution, when a certain value in a set of data... If the following condition is met, then it is considered that... These are outlier data and should be removed from the data set. The removed data should then be filled with the mean of the data set.

[0064] (1)

[0065] In the formula This is suspicious data to be examined. for The mean of a set of data. This is the critical value for the Grubbs test. At the significance level, The number of samples in each data set. is the standard deviation of this data set. The significance level is... = 0.05, number of samples in each data group = 20, the corresponding Grubbs critical value is = 2.557.

[0066] Because the amount of fertilizer applied by the fertilizer applicator is highly sensitive to changes in the opening when it is operating at a small opening, and the growth rate of fertilizer application varies drastically, the initial data is segmented to avoid the difference in fertilizer growth rate affecting the accuracy of the fertilizer rate prediction model.

[0067] First, the average growth rate of the fertilization rate corresponding to each opening degree and its previous opening degree is calculated, specifically expressed as follows:

[0068] (2)

[0069] In the formula, For the first The average fertilization growth rate corresponding to each opening degree compared to the previous opening degree. For the first The fertilization rate corresponding to each opening degree and the j-th rotation speed This represents the fertilization rate corresponding to the previous opening degree and the j-th rotation speed. This represents the total number of rotational speeds.

[0070] The average growth rate of the above fertilization rates Group the data and find the maximum and minimum variances of all possible groups. Use the median of the maximum and minimum variances as the segment reference values. Take the data group whose variance differs the least from the reference value as segment A, and the remaining data as segment B.

[0071] 2. Training of the fertilizer application prediction model.

[0072] The flowchart of the fertilizer application prediction model in the bivariate fertilizer application decision-making method for fertilizer applicators provided in this embodiment is as follows: Figure 3 As shown, for the segmented data, the dataset is first divided into training, validation, and test sets for each segment. The training set accounts for 72% of each segment, the validation set for 18%, and the test set for 10%. To avoid disrupting the original distribution of the data during dataset splitting, each segment is first divided into... A small interval, This represents the number of samples in each data segment. One sample is drawn from each segment as the test set. The validation set is partitioned using the same method. The datasets after each data segment is partitioned are then normalized to avoid the impact of different units of measurement on the model's prediction accuracy during training.

[0073] This embodiment provides a training process for a fertilizer application prediction model based on an improved extreme learning machine, as follows: Figure 3 As shown, the improved extreme learning machine consists of an input layer, a hidden layer, and an output layer. The rotational speed N and aperture L from the two segments of data after the above segmentation are used as inputs, and the fertilizer application rate Q is used as the output. It is assumed that... Then the improved Extreme Learning Machine The model is defined as follows:

[0074] (3)

[0075] In the formula, For the first A combination of rotational speed and opening degree For the first The amount of fertilizer applied corresponding to each rotation speed and opening degree. For the input layer and the first The connection weights of each hidden layer neuron For the first The threshold of a hidden layer neuron For the first The connection weights between hidden layer neurons and the output layer For the first One activation function, For the first The weights of each activation function, and The number of neurons in the hidden layer is calculated based on empirical formulas, and then randomly selected... and Perform the assignment.

[0076] To simplify the above N expressions, they can be written in matrix form, specifically as follows:

[0077] (4)

[0078] In the formula, H is the hidden layer output matrix. ,in It is a linear combination of multiple activation functions, specifically expressed as: , To input the number of samples, This represents the number of neurons in the hidden layer.

[0079] because and The values ​​are generated randomly, so during model training, only the connection weights between the hidden layer and the output layer need to be calculated. To ensure the minimum training error, Let the objective function be calculated using the least squares method. The optimal value: ,in For matrix The generalized inverse, and .

[0080] Since the weight differences of different activation functions in a multi-core extreme learning machine can affect the prediction performance of the final trained model, an optimal set of activation function weight vectors should be selected. In this embodiment, an adaptive compression factor particle swarm optimization algorithm is used to optimize the weight vector with the goal of minimizing the mean squared error of the model on the validation set. The speed update formula of the adaptive compression factor particle swarm optimization algorithm is as follows:

[0081] (5)

[0082] In the formula, the compressibility factor ,

[0083] The learning factor is dynamically adjusted based on the change in the number of iterations. and This is to improve the algorithm's global search capability in the early stages of iteration and its local search capability in the later stages of iteration, and based on... , For compressibility factor Dynamic adjustments are made to further control particle velocity, balancing the algorithm's global and local search capabilities. , The update formula is specifically expressed as follows:

[0084] (6)

[0085] (7)

[0086] In the formula, , This represents the current iteration number. To minimize the number of iterations, This represents the maximum number of iterations.

[0087] The particle position update formula is as follows:

[0088] (8)

[0089] The fitness function is defined as follows:

[0090] (9)

[0091] In the formula, The activation function weight vector, This refers to the actual amount of fertilizer applied. To predict the amount of fertilizer to apply, This represents the number of samples in the validation set.

[0092] After determining the above steps, set the basic parameters of the particle swarm optimization algorithm, including population size Pop, particle dimension D, and learning factor. and Maximum number of iterations T.

[0093] During the optimization of the activation function weight vector, the particle velocity is updated by Equation (5), the particle position is updated by Equation (8), and then the fitness value of the current particle position is calculated and recorded by Equation (9) until the maximum number of iterations is met. The minimum value among all fitness values ​​and its corresponding particle are found. This particle is the optimal activation function weight vector.

[0094] After the model is trained, the model is used to predict the amount of fertilizer applied corresponding to the rotation speed and opening degree in the test set, and the mean square error between the predicted amount of fertilizer and the actual amount of fertilizer applied in the test set is calculated to evaluate the prediction accuracy of the model.

[0095] 3. Fertilization decision-making solution.

[0096] This embodiment provides a bivariate fertilizer application decision-making method for fertilizer applicators, as follows: Figure 4 As shown, a multi-objective fertilization performance problem model is first established, including optimization solutions for fertilization accuracy, fertilization uniformity, equipment adjustment timeliness, and fertilizer breakage rate. The fertilization accuracy objective model is defined as minimizing the absolute value of the squared error between the actual required fertilization amount from the fertilization prescription map and the predicted value obtained from the fertilization amount prediction model, which is equal to the maximum permissible error of the fertilizer applicator. The fertilization accuracy problem model is specifically expressed as follows:

[0097] (10)

[0098] In the formula, The rotational speed of the fertilizer discharge shaft. Q represents the opening of the feed inlet, and Q represents the actual amount of fertilizer required according to the fertilizer prescription diagram. The predicted value is from the fertilizer application rate prediction model. This represents the maximum permissible error for the fertilizer applicator.

[0099] Considering that the fertilization uniformity of a dual-variable fertilizer applicator improves with increasing discharge shaft speed, and since a fixed fertilization rate may correspond to multiple different speeds and openings, to ensure good fertilization uniformity, the larger value among the speeds corresponding to the target fertilization rate should be chosen. Furthermore, for a fixed fertilization rate, the angle formed by the line connecting different combinations of speeds and openings to the origin and the x-axis in a rectangular coordinate system varies. Assuming the x-axis represents the discharge port opening and the y-axis represents the discharge shaft speed, this angle increases with increasing speed and decreases with increasing opening. Therefore, the fertilization uniformity can be evaluated by transforming the speed and opening into rectangular and polar coordinates, and by using the polar angle in the polar coordinate system. Since fertilization uniformity improves with increasing speed, and increasing speed increases the polar angle, the fertilization uniformity problem model for a target fertilization rate is defined as the maximum value of the polar angle. To standardize the problem model, the minimum value is used. Therefore, the target fertilization uniformity model can be defined as:

[0100] = (11)

[0101] In the formula This represents the polar angle of the rotational speed and opening in polar coordinates.

[0102] The equipment adjustment time mainly stems from the time spent due to mechanical lag when the target fertilizer application rate changes. To minimize this time, the adjustment range of the rotation speed and opening corresponding to two changes in fertilizer application rate should be minimized. Therefore, the target model for equipment adjustment timeliness can be defined as minimizing the absolute value of the difference between the polar angles of the rotation speed and opening corresponding to two fertilizer application rates in polar coordinates, specifically expressed as:

[0103] = (12)

[0104] In the formula, This represents the polar angle of the rotation speed opening corresponding to the current fertilization rate in polar coordinates. This represents the polar angle of the rotation speed opening corresponding to the previous fertilization rate in polar coordinates.

[0105] Fertilizer breakage rate increases with rotational speed. To determine the specific trend of fertilizer breakage rate with rotational speed, experiments were conducted to test the breakage rate at rotational speeds of 150 r / min, 125 r / min, and 100 r / min. The results showed that the breakage rate was highest at 150 r / min, with an average value of 3.5%; the average value was 1.87% at 125 r / min; and the average value was 0.69% at 100 r / min. To reduce the breakage rate while ensuring fertilization uniformity as much as possible, the rotational speed should be kept as close to 100 r / min as possible. Since the opening has little impact on the breakage rate, to simultaneously obtain a shorter equipment adjustment time, a secondary constraint can be applied to the opening while optimizing the breakage rate, ensuring that the change in opening is as small as possible relative to the previous fertilization value. The target model for fertilizer breakage rate is defined as follows:

[0106] = (13)

[0107] In the formula, This represents the polar angle of the rotation speed opening corresponding to the current fertilization rate in polar coordinates. The corresponding rotation speed is 100 r / min, and the opening degree is the polar angle of the opening degree corresponding to the previous fertilization in polar coordinates.

[0108] After the models for each objective problem are defined, the improved NSGA-III is used to solve the model. The solution process can be roughly divided into: reference point setting, population adaptive normalization, associating reference points and individual retention.

[0109] Reference points are set using a boundary intersection weighting method, assuming that the hyperplane boundary corresponding to each target is uniformly divided into... If the portion is... The total number of reference points H corresponding to each target is .

[0110] After the reference point is set, the population is initialized to generate the first generation population. For the subsequent t-th generation population, an archive set is established. Genetic algorithms are used to determine the parent population. Generate offspring population ,merge and Generate population Populations are sorted by non-dominated ordering. The system is divided into different levels of non-dominated layers, and individuals with higher non-dominated layers are stored in the appropriate layers. In the middle, if (Population size), then ,like Then it is necessary to The last non-dominated layer undergoes adaptive normalization, reference point association, and individual preservation operations until... .

[0111] Population adaptive normalization selects the minimum value of each objective to form the ideal point of the population. The population is determined by equation (14). The translation transforms the ideal point into a zero vector.

[0112] (14)

[0113] In the formula, For the first Target value, For the first The ideal value of the target For the translation of the first Target value.

[0114] The extreme points corresponding to each target are obtained through equation (15), specifically expressed as follows:

[0115] (15)

[0116] In the formula, Standardize the direction vectors of the coordinate axes, when At that time, through replace.

[0117] By connecting the extreme points corresponding to the number M of fertilization performance targets, an M-dimensional linear hyperplane is formed. At this point, the th... The intercepts of the coordinate axes corresponding to each target with the M-dimensional linear hyperplane Then the normalization formula for the objective function is:

[0118] = = , (16)

[0119] in, .

[0120] After the population is normalized, reference points are associated with individuals in the population. In the original NSGA-III, the line connecting the origin to the reference point is first defined as the reference line. The reference point association operation is performed by calculating the shortest vertical distance from the individual in the population to the reference line. This method performs well in terms of solution diversity, but is not good in terms of solution convergence.

[0121] This embodiment provides a bivariate fertilization decision-making method for a fertilizer applicator. First, a connecting line is formed between individuals in the population and a reference point. At this point, the connecting line and the reference point form an angle. To ensure both convergence and diversity of solutions, the angle between the individual-reference point line and the reference line is calculated and selected. The reference point is associated with the point where the sum of the sine and cosine values ​​is minimized.

[0122] After the reference point association is completed, record the population. The number of individuals associated with each reference point, assuming the number of individuals associated with the j-th reference point is... If the reference point has the fewest associated individuals, then its corresponding individuals will be retained first. If there are multiple reference points, then one of them will be randomly selected. The corresponding individuals are retained.

[0123] like = 0, which indicates a population There are no individuals or reference points in it. Association, at this time in the population Search for individuals associated with it in the population. If an individual exists in the population that is associated with the reference point, then... For individuals that are associated, the individual closest to the reference line corresponding to that reference point will be added to the population. In the middle, at this time If the population If no individual is associated with this reference point, the remaining steps for that reference point will not be considered. , indicating population If an individual already exists in the population associated with the reference point, then randomly select one from the population that is associated with the reference point. Related individuals join the population Repeat the above steps until... The size is equal to N.

[0124] When the program satisfies the given maximum number of iterations T, the solution process ends and outputs the optimal solution set of the four objective problems corresponding to the target fertilization amount and the corresponding optimal fertilization decision. When carrying out fertilization operations, the appropriate fertilization decision can be selected from the solution set according to actual needs.

Claims

1. A bivariate fertilization decision-making method for fertilizer applicators, characterized in that, Includes the following steps: 1) Prepare training data; the training data includes several sets of fertilizer discharge shaft speed and discharge port opening of the fertilizer applicator as model inputs, and the corresponding fertilizer application amount of the fertilizer applicator as model outputs; 2) The improved extreme learning machine model is trained using training data to obtain a fertilizer application prediction model; 3) Establish a fertilization performance model with one or more of the following as objectives: highest fertilization accuracy, highest fertilization uniformity, highest equipment adjustment timeliness, and lowest fertilizer breakage rate. Solve the fertilization performance model with different combinations of feed port opening and fertilizer discharge shaft speed as constraints to obtain the optimal solution set of the speed and opening combination corresponding to the target fertilizer amount. The target fertilization rate is obtained based on the fertilization rate prediction model, and the fertilization performance model is solved using an improved NSGA-III. The improved NSGA-III refers to an improvement on the reference point association operation based on NSGA-III. The improved reference point association operation is as follows: when the sum of the sine and cosine values ​​of the angle between the line connecting the individual to the reference point and the reference line is minimized, the individual is associated with the reference point. The target for the highest fertilization accuracy is defined as having the smallest absolute difference between the fertilization rate prediction value of the fertilization rate prediction model and the actual fertilization rate required by the fertilization prescription map; fertilization uniformity and equipment adjustment are also considered. The optimization target model for timeliness and fertilizer breakage rate is defined by transforming rotation speed and opening degree from rectangular coordinates to polar coordinates. In rectangular coordinates, the horizontal axis represents the opening degree, and the vertical axis represents the rotation speed. The target for the highest fertilizer uniformity is defined as the largest polar angle in polar coordinates. The target for the highest equipment adjustment timeliness is defined as the smallest absolute value of the difference between the polar angles of the rotation speed and opening degree corresponding to two fertilization rates in polar coordinates. The target for the lowest fertilizer breakage rate is defined as the smallest absolute value of the difference between the polar angle of the current fertilization rate corresponding to the rotation speed and opening degree in polar coordinates and the polar angle of the rotation speed of 100 r / min and the opening degree corresponding to the previous fertilization rate in polar coordinates. 4) During fertilization, select the combination of rotation speed and opening that satisfies the fertilizer application requirements from the optimal solution set as the fertilizer application decision control fertilizer applicator.

2. The bivariate fertilization decision-making method for fertilizer applicators according to claim 1, characterized in that, The preparation of the training data includes the following preprocessing steps: initial data on fertilizer application amount are collected using a fertilizer applicator frame with different combinations of rotation speed and opening, and then outlier detection and removal are performed on the initial data. The data after outlier removal is segmented according to the degree of influence of the feed opening on the fertilizer growth trend. Each segment of data is used as the training set, validation set and test set for training the extreme learning machine model.

3. The bivariate fertilization decision-making method for fertilizer applicators according to claim 2, characterized in that, Outliers in the initial data are detected using the Grubbs criterion, and the outliers are filled in by the mean of the data set.

4. The bivariate fertilization decision-making method for fertilizer applicators according to claim 2, characterized in that, The data is segmented as follows: First, the average growth rate of the fertilization rate corresponding to each feed inlet opening degree and its previous opening degree is calculated, specifically expressed as: In the formula, For the first The average fertilization growth rate corresponding to each opening degree compared to the previous opening degree. For the first The fertilization rate corresponding to each opening degree and the j-th rotation speed This represents the fertilization rate corresponding to the previous opening degree and the j-th rotation speed. The total number of rotational speeds; This leads to the average growth rate of fertilization rate. The average growth rate of fertilization rate Group the data and find the maximum and minimum variances of all possible groups. Use the median of the maximum and minimum variances as the segment reference values. Take the data group whose variance differs the least from the segment reference value as one segment and the remaining data as another segment.

5. The bivariate fertilization decision-making method for fertilizer applicators according to claim 1, characterized in that, The improved extreme learning machine model yields fertilizer application prediction models, including: The construction of a multi-core single-hidden-layer extreme learning machine model is specifically represented as follows: in, The connection weights between the hidden layer and the output layer. For the first One activation function, The weights of the activation function, , These are the connection weights between neurons in the input layer and the hidden layer. The threshold of the hidden layer neurons. For the input sample, This represents the number of neurons in the hidden layer. The number of activation functions. For output samples; In training the multi-core single-hidden-layer extreme learning machine model, each segment of the preprocessed training data is divided into a training set, a validation set, and a test set. During the segmentation, each segment of data is divided into several small intervals, and one sample is extracted from each small interval as the data for the test set and the validation set. The preprocessed training data is also normalized to reduce the fertilizer discharge shaft speed, the discharge port opening, and the corresponding fertilizer application rate to between 0 and 1.

6. The bivariate fertilization decision-making method for fertilizer applicators according to claim 5, characterized in that, An adaptive compression factor particle swarm optimization algorithm is used to minimize the mean squared error of the model on the validation set, adjusting the activation function weights. Optimization is performed to obtain the optimal activation function weight vector.

7. The bivariate fertilization decision-making method for fertilizer applicators according to claim 6, characterized in that, The velocity update formula for the adaptive compression factor particle swarm is: Among them, compression factor , , and For learning factors.

8. The bivariate fertilization decision-making method for fertilizer applicators according to claim 7, characterized in that, The adaptive compression factor particle swarm algorithm dynamically adjusts the learning factor based on the change in the number of iterations. and and through , For compressibility factor Make dynamic adjustments. , The updated formula is: in, , This represents the current iteration number. To minimize the number of iterations, This represents the maximum number of iterations.

9. The bivariate fertilization decision-making method for fertilizer applicators according to claim 1, characterized in that, The highest accuracy of fertilization is specifically expressed as: in, The rotational speed of the fertilizer discharge shaft. For the opening of the feed inlet, The required amount of fertilizer, This is the predicted value for fertilizer application. This represents the maximum permissible error for the fertilizer applicator. The highest level of fertilization uniformity is specifically expressed as: = in, This represents the polar angle of the rotational speed and opening in polar coordinates.

10. The bivariate fertilization decision-making method for fertilizer applicators according to claim 8, characterized in that, The fitness function is: In the formula, The activation function weight vector, This refers to the actual amount of fertilizer applied. To predict the amount of fertilizer to apply, This represents the number of samples in the validation set.

Images