A method for constructing a prediction model for non-grain soil acidification

Abstract

The present invention discloses a method for constructing a prediction model for non-grain soil acidification, comprising the following steps: S1, collecting soil sample data of various types of non-grain land in different regions; S2, preprocessing the collected soil sample data; S3, feature selection and extraction; S4, constructing a model using a machine learning algorithm; S5, model training and optimization; S6, evaluating the trained model using a test set. The model construction method of the present invention can comprehensively consider the influence of various factors on "non-grain" soil acidification, and the prediction result is more accurate and reliable compared with the traditional single factor model or simple empirical model. The present invention can effectively mine the complex relationship in the data through the automatic learning and feature selection of the machine learning algorithm, adapt to the soil acidification prediction needs of different regions and different types of "non-grain" land, and has strong versatility and adaptability.

Description

Technical Field

[0001] The present invention relates to the technical field of soil environment monitoring and prediction, and more specifically, to a method for constructing a soil acidification prediction model caused by non-grain land use changes. Background Art

[0002] With economic development and the adjustment of agricultural industrial structure, the phenomenon of "non-grainization" has changed the land use pattern to a certain extent. "Non-grainization" means that the arable land originally used for grain production is used for planting cash crops (such as flowers, fruits, vegetables, etc.), developing forestry, breeding and other non-grain production activities. This change in land use has many impacts on soil properties, among which soil acidification has received increasing attention.

[0003] Soil acidification can lead to a decrease in soil fertility, affect the activity of microorganisms in the soil, and further affect the growth and yield of crops. Traditional soil monitoring methods mainly rely on field sampling and laboratory analysis. Although this method is accurate, it has problems such as high cost, time lag, and limited sample points, and it is difficult to meet the needs of large-scale, real-time monitoring and prediction. Therefore, it is of great significance to establish an efficient and accurate "non-grain" soil acidification prediction model.

[0004] At present, although there are some methods for constructing soil property prediction models, there are still few prediction models specifically for soil acidification caused by "non-grainization". Existing soil acidification prediction models often do not fully consider the combined effects of different "non-grainization" types, different soil textures, and various environmental factors (such as climate, fertilization management, etc.), resulting in low accuracy of prediction results. Summary of the invention

[0005] The present invention aims to provide a method for constructing a "non-grain" soil acidification prediction model, which can comprehensively consider multiple factors and accurately predict the trend of soil acidification under "non-grain" land use changes, providing a scientific basis for soil management and sustainable agricultural development.

[0006] In order to achieve these purposes and other advantages according to the present invention, on the one hand, a preferred embodiment of the present invention provides a method for constructing a non-grain soil acidification prediction model, comprising the following steps:

[0007] S1. Data Collection

[0008] Collect soil sample data from different regions and various types of non-grain land;

[0009] S2. Data preprocessing

[0010] Preprocess the collected soil sample data;

[0011] S3. Feature selection and extraction

[0012] Analyze the correlation between the characteristics of various factors in soil sample data and soil acidification, select the factors with significant impact on soil acidification as model input variables, and encode and quantify the selected factors;

[0013] S4. Model construction

[0014] Build models using machine learning algorithms;

[0015] S5. Model training and optimization

[0016] The data set is divided into a training set and a test set. The constructed model is trained using the training set. The model performance is optimized by adjusting the model parameters. The cross-validation method is used to select the parameter combination with the best performance on the validation set. The loss function is used during the training process to measure the difference between the model prediction result and the true value, and the model parameters are updated by minimizing the loss function through the gradient descent optimization algorithm.

[0017] S6. Model evaluation and application

[0018] Use the test set to evaluate the trained model and calculate the evaluation indicators. If the model evaluation indicators meet the requirements, the model will be applied to the actual non-grain soil acidification prediction, and new non-grain land-related factor characteristic data will be input. The model will output the soil acidification prediction results.

[0019] Preferably, the non-grain land types include flower planting areas, orchards, woodlands, and breeding grounds; the climate data include the average annual rainfall, average temperature, and sunshine hours; and the fertilization management data include the type of fertilizer, the amount of fertilizer, and the frequency of fertilization.

[0020] Preferably, in S1, the soil sample data includes soil pH value, soil texture, soil organic matter content, soil cation exchange capacity, land use history, climate data and fertilization management data.

[0021] Preferably, in S2, the collected soil sample data is preprocessed, specifically including the following operations:

[0022] For missing data on soil organic matter content, estimates are made based on land use types and similar soil conditions in the surrounding area, or data interpolation methods are used to supplement them. For abnormal values, judgments and corrections are made by comparing them with the distribution of data on the same type of land to obtain normal values.

[0023] Preferably, the S3 specifically includes the following steps:

[0024] The Pearson correlation coefficient between soil pH and other factors was calculated to screen the factor characteristics, the categorical variables were converted into numerical variables using one-hot encoding, and the continuous variables were standardized.

[0025] Preferably, in S4, when the random forest algorithm is used to construct the model, the random forest consists of multiple decision trees, the number of decision trees is determined, samples are randomly extracted with replacement from the training data set to construct the decision tree, a part of the factor features are randomly selected from all factor features when the node is split, the best splitting factor feature is selected according to the information gain, information gain ratio or Gini index, the decision tree grows until the stopping condition is met, and the prediction results of multiple decision trees are integrated through the idea of ​​ensemble learning.

[0026] Preferably, in S5, the data set is divided into 70% for training set and 30% for test set, 10-fold cross validation is adopted, the mean square error function is used as the loss function for regression problem, the cross entropy function is used as the loss function for classification problem, and the model parameters are updated by stochastic gradient descent optimization algorithm.

[0027] The present invention has at least the following beneficial effects:

[0028] (1) The model construction method of the present invention can comprehensively consider the impact of multiple factors on "non-grain" soil acidification. Compared with traditional single-factor models or simple empirical models, the prediction results are more accurate and reliable.

[0029] (2) Through the automatic learning and factor feature selection of machine learning algorithms, the present invention can effectively mine the complex relationships in the data and adapt to the soil acidification prediction needs of different regions and different types of "non-grain" lands, and has strong versatility and adaptability.

[0030] (3) The prediction results based on the model of the present invention can provide a scientific basis for agricultural producers and land managers to take measures in advance to prevent and improve soil acidification problems, which is conducive to the sustainable development of agriculture and the rational use of land resources.

[0031] Other advantages, objectives and features of the present invention will be embodied in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic diagram of the steps of the method for constructing the non-grain soil acidification prediction model in the present invention. DETAILED DESCRIPTION

[0033] The present invention is further described in detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0034] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0035] like Figure 1 As shown, on the one hand, a preferred embodiment of the present invention provides a method for constructing a non-grain soil acidification prediction model, comprising the following steps:

[0036] S1. Data Collection

[0037] Collect soil sample data from different regions and various types of non-grain land;

[0038] In S1, the non-grain land types include flower planting areas, orchards, woodlands, and breeding grounds; the climate data include the average annual rainfall, average temperature, and sunshine hours; and the fertilization management data include the type of fertilizer, the amount of fertilizer, and the frequency of fertilization.

[0039] The soil sample data includes soil pH value, soil texture, soil organic matter content, soil cation exchange capacity, land use history, climate data and fertilization management data.

[0040] S2. Data preprocessing

[0041] Preprocess the collected soil sample data;

[0042] In S2, the collected soil sample data is preprocessed, specifically including the following operations:

[0043] For missing data on soil organic matter content, estimates are made based on land use types and similar soil conditions in the surrounding area, or data interpolation methods are used to supplement them. For abnormal values, judgments and corrections are made by comparing them with the distribution of data on the same type of land to obtain normal values.

[0044] S3. Feature selection and extraction

[0045] Analyze the correlation between each factor characteristic and soil acidification, select the factor characteristics with significant impact on soil acidification as the model input variables, and encode and quantify the selected factor characteristics;

[0046] The S3 specifically includes the following steps:

[0047] The Pearson correlation coefficient between soil pH and other factors was calculated to screen the factor characteristics, the categorical variables were converted into numerical variables using one-hot encoding, and the continuous variables were standardized.

[0048] Among them, the factor characteristics were screened by calculating the Pearson correlation coefficient between soil pH value and other factors (corresponding to: analyzing the correlation between the characteristics of each factor in the soil sample data and soil acidification), the categorical variables were converted into numerical variables using one-hot encoding (corresponding to: screening out the factor characteristics that significantly affect soil acidification as model input variables, and encoding the selected factor characteristics), and the continuous variables were standardized (corresponding to: quantification processing).

[0049] S4. Model construction

[0050] Build models using machine learning algorithms;

[0051] In S4, when the random forest algorithm is used to construct the model, the random forest is composed of multiple decision trees, the number of decision trees is determined, samples are randomly selected with replacement from the training data set to construct the decision tree, a part of the factor features are randomly selected from all factor features when the node is split, the best split feature is selected according to the information gain, information gain ratio or Gini index, the decision tree grows until the stopping condition is met, and the prediction results of multiple decision trees are integrated through the ensemble learning idea

[0052] S5. Model training and optimization

[0053] The data set is divided into a training set and a test set. The constructed model is trained using the training set. The model performance is optimized by adjusting the model parameters. The cross-validation method is used to select the parameter combination with the best performance on the validation set. The loss function is used during the training process to measure the difference between the model prediction result and the true value, and the model parameters are updated by minimizing the loss function through the gradient descent optimization algorithm.

[0054] In S5, the data set is divided into 70% for training set and 30% for testing set, 10-fold cross validation is adopted, the mean square error function is used as the loss function for regression problem, the cross entropy function is used as the loss function for classification problem, and the model parameters are updated by stochastic gradient descent optimization algorithm.

[0055] S6. Model evaluation and application

[0056] Use the test set to evaluate the trained model and calculate the evaluation indicators. If the model evaluation indicators meet the requirements, the model will be applied to the actual non-grain soil acidification prediction, and new non-grain land-related factor characteristic data will be input. The model will output the soil acidification prediction results.

[0057] To facilitate understanding of the technology, a specific implementation is provided below.

[0058] A preferred embodiment of the present invention provides a method for constructing a non-grain soil acidification prediction model, comprising the following steps:

[0059] S1. Data collection steps:

[0060] A professional soil sampling team was organized to select several representative "non-grain" land monitoring areas across the country, covering different climate zones (such as tropical, subtropical, temperate, etc.), topography (such as plains, mountains, hills, etc.) and soil types (such as red soil, yellow soil, black soil, brown soil, etc.). In each monitoring area, multiple sampling points were set up for different "non-grain" types (such as flower planting areas, orchards, tea gardens, woodlands, breeding areas, etc.).

[0061] For each sampling point, the land use history information was recorded in detail, including the types of food crops previously planted (such as rice, wheat, corn, etc.), the planting years (accurate to the year), and the specific time of conversion to the current "non-food" use (accurate to the month). At the same time, relevant information such as local land management policies and land transfer conditions was collected to comprehensively analyze the impact of land use changes on soil acidification.

[0062] Use professional soil sampling equipment to collect soil samples according to the stratified sampling method. At each sampling point, soil samples are collected at different depths such as 0-10 cm, 10-20 cm, 20-30 cm and 30-40 cm. Multiple sub-samples (such as 3-5) are collected at each depth, and then the sub-samples at the same depth are mixed evenly as representative samples of that depth. In the laboratory, various indicators of soil samples are determined strictly in accordance with standard analytical methods. The soil pH value is measured using the glass electrode method and a high-precision pH meter to ensure that the measurement accuracy reaches 0.01 pH units; the soil texture is determined by combining the screening method and the hydrometer method to determine the content ratio of sand, loam and clay; the soil organic matter content is determined by potassium dichromate oxidation-external heating method; the soil cation exchange capacity is determined by the ammonium acetate exchange method.

[0063] Obtain detailed climate data for the past 20 years in each monitoring area from the national meteorological department, including annual average rainfall (accurate to millimeters), monthly average rainfall (accurate to millimeters), annual average temperature (accurate to 0.1 degrees Celsius), monthly average temperature (accurate to 0.1 degrees Celsius), sunshine hours (accurate to hours), precipitation days (accurate to days), average relative humidity (accurate to 1%), etc. At the same time, cooperate with local agricultural technology extension stations, farmers and related agricultural enterprises to collect fertilization management data. Record detailed information such as the type of fertilizer used (such as organic fertilizer, chemical fertilizer, compound fertilizer, etc.), the content of each major nutrient (nitrogen, phosphorus, potassium, etc.) in the fertilizer (accurate to 1%), the number of fertilizations per year (accurate to times), the time of each fertilization (accurate to days) and the amount of fertilizer applied (accurate to kilograms per mu). For irrigation conditions, record the irrigation water source (such as river water, well water, rainwater collection, etc.), irrigation method (such as flood irrigation, drip irrigation, sprinkler irrigation, etc.), irrigation frequency (such as several times a week, several times a month, etc.) and the amount of water for each irrigation (accurate to cubic meters / mu).

[0064] S2. Data preprocessing steps:

[0065] Perform comprehensive preprocessing on the collected data.

[0066] For missing soil data, such as missing soil organic matter content data at some sampling points, estimates are made based on land use type, soil texture, and organic matter content data at surrounding sampling points. A distance-weighted interpolation method can be used, that is, different weights are assigned according to the distance from surrounding known sampling points to calculate the estimated value of the missing points. For outliers, they are judged by comparing and analyzing the data of sampling points on the same type of "non-grain" land, the same soil type, and similar climatic conditions. For example, if the soil pH value of a certain sampling point deviates significantly from the pH value range of other similar sampling points (such as exceeding the mean ±3 times the standard deviation), check whether there is an error in the measurement process. If it is determined to be an outlier, it can be corrected based on the data of surrounding sampling points, or if there is sufficient data, consider removing the outlier.

[0067] S3, feature selection and extraction steps:

[0068] Calculate the Pearson correlation coefficient between soil pH and other collected factors (such as soil texture component content, soil organic matter content, soil cation exchange capacity, climate data, fertilization management data, etc.). The calculation formula is:

[0069]

[0070] where x i and i are the i-th observed values ​​of two variables (such as soil pH and soil organic matter content), and is their average value, and n is the number of observations. By calculating the correlation coefficient, factors with an absolute correlation value greater than 0.3 (this threshold can be adjusted according to actual conditions) are selected as input features of the model. For example, it is found that factors such as clay content in soil texture, soil organic matter content, annual average rainfall, and nitrogen fertilizer application in fertilization have a high correlation with soil pH.

[0071] The selected features are encoded and quantified. For categorical variables such as soil texture (sand, loam, clay), one-hot encoding is used. For example, if there are three categories of sand, loam, and clay, when the soil texture is sand, it is encoded as (1,0,0); when it is loam, it is encoded as (0,1,0); when it is clay, it is encoded as (0,0,1). For continuous variables such as soil organic matter content and fertilizer application, standardization is performed. The z-score standardization method is used, and the calculation formula is: Where x is the original value, μ is the mean of the feature, and σ is the standard deviation. After standardization, the range of the feature value is roughly between -3 and 3 (under the normal distribution assumption, about 99.7% of the data will fall within this range), which is convenient for model calculation and parameter optimization.

[0072] S4. Model construction steps:

[0073] The random forest algorithm is selected to build the prediction model. Random forest is an ensemble learning algorithm based on decision trees, which has good generalization ability and anti-overfitting performance.

[0074] First, determine the number of decision trees n. Select a suitable value by comparing the effects of different numbers of decision trees on model performance through experiments. For example, start from 50 decision trees and gradually increase to 200 trees, increasing by 10 trees each time, and evaluate the model performance on the training set and validation set (such as calculating mean square error, accuracy, and other indicators). It is found that when the number of decision trees reaches 150, the model performance on the validation set tends to be stable, and further increasing the number of decision trees does not significantly improve the performance, so m = 150 is selected.

[0075] For each decision tree, random sampling with replacement is performed from the training data set. For example, if the training data set has m samples, m samples (possibly repeated samples) are randomly selected each time as the training data for the decision tree. This process is called bootstrap sampling. In this way, the training data of each decision tree has a certain degree of randomness, which increases the diversity of the model.

[0076] When a node splits, a portion of features is randomly selected from all features. Let the total number of features be p, and select features (this is a common method for selecting the number of features in the random forest algorithm). Then the best split feature is selected based on indicators such as information gain, information gain ratio or Gini index. The calculation formula for information gain is:

[0077]

[0078] Where D is the data set of the current node, a is the feature to be split, V is the number of values ​​of feature a, Ent(D) is the information entropy of the data set, and Ent(D v ) is the information entropy of the subset when feature a takes the value v. By calculating the information gain, the feature that maximizes the information gain is selected as the node splitting feature to build a decision tree. Each decision tree continues to grow until the stopping condition is met. The stopping condition can be set as the number of node samples is less than a certain threshold (such as 10 samples), or the depth of the tree reaches a certain limit (such as a maximum depth of 10 layers) to prevent the decision tree from overfitting.

[0079] Through the idea of ​​ensemble learning, the prediction results of multiple decision trees are integrated. For regression problems (such as predicting the change in soil pH), the average method is used. That is, for a new sample, the prediction values ​​of 150 decision trees are averaged as the final prediction value. For classification problems (such as predicting whether the soil is seriously acidified, it can be set that the pH value is less than a certain threshold as severe acidification, otherwise it is not severe), the voting method is used. That is, the number of severe and non-serious acidification predictions in the 150 decision trees is counted, and the category with the largest number is taken as the final prediction result.

[0080] S5, model training and optimization steps:

[0081] The sorted data set is divided into a training set and a test set according to a certain ratio (such as 70% for training set and 30% for test set).

[0082] The constructed model is trained using the training set, and the model performance is optimized by adjusting the model parameters (such as the number of decision trees in the random forest, the depth of the decision tree, the number of selected features when splitting nodes, etc.). The training set is further divided into a training subset and a validation subset using a cross-validation method (such as 10-fold cross-validation). In each cross-validation, the model is trained using the training subset, and the model performance is evaluated using the validation subset (such as calculating indicators such as mean square error and accuracy), and the parameter combination with the best performance on the validation set is selected.

[0083] During the training process, a loss function (such as the mean square error function for regression problems and the cross entropy function for classification problems) is used to measure the difference between the model prediction results and the true value, and optimization algorithms such as gradient descent (such as stochastic gradient descent, SGD) are used to minimize the loss function, continuously update the model parameters, and improve the accuracy of the model.

[0084] S6. Model evaluation and application:

[0085] Use the test set to comprehensively evaluate the trained model and calculate the evaluation indicators (such as the determination coefficient R2, root mean square error RMSE, etc. for regression models; accuracy, recall rate, F1 value, etc. for classification models). For example, the closer the determination coefficient R2 is to 1, the better the model fit is; the smaller the root mean square error RMSE is, the closer the model prediction value is to the true value.

[0086] If the model evaluation indicators meet the predetermined requirements, the model can be applied to the actual "non-grain" soil acidification prediction. Input the relevant characteristic data of the new "non-grain" land (such as soil texture, climate conditions, fertilization management, etc.), and the model outputs the prediction results of soil acidification (such as the changing trend of soil pH value, whether severe acidification is likely to occur, etc.). Based on the prediction results, scientific and reasonable decision-making support is provided for soil improvement, agricultural planting planning, etc. For example, if it is predicted that the soil acidification trend of a flower planting area is obvious, improvement measures can be taken in advance, such as applying lime to increase soil pH value, optimizing fertilization schemes, etc.

[0087] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes, and they can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

Claims

1. A method for constructing a non-grain soil acidification prediction model, characterized in that: The following steps are involved: S1. Data Collection Collect soil sample data from different regions and various types of non-grain land; S2. Data preprocessing Preprocess the collected soil sample data; S3. Feature selection and extraction Analyze the correlation between the characteristics of various factors in soil sample data and soil acidification, select the factors with significant impact on soil acidification as model input variables, and encode and quantify the selected factors; S4. Model construction Build models using machine learning algorithms; S5. Model training and optimization The data set is divided into a training set and a test set. The constructed model is trained using the training set. The model performance is optimized by adjusting the model parameters. The cross-validation method is used to select the parameter combination with the best performance on the validation set. The loss function is used during the training process to measure the difference between the model prediction result and the true value, and the model parameters are updated by minimizing the loss function through the gradient descent optimization algorithm. S6. Model evaluation and application Use the test set to evaluate the trained model and calculate the evaluation indicators. If the model evaluation indicators meet the requirements, the model will be applied to the actual non-grain soil acidification prediction, and new non-grain land-related factor characteristic data will be input. The model will output the soil acidification prediction results.

2. The method for constructing a non-grain soil acidification prediction model according to claim 1, characterized in that: The non-grain land types include flower plantations, orchards, woodlands, and breeding grounds; the climate data include average annual rainfall, average temperature, and sunshine hours; and the fertilization management data include fertilizer type, fertilizer amount, and fertilization frequency.

3. The method for constructing a non-grain soil acidification prediction model according to claim 2, characterized in that: In S1, the soil sample data includes soil pH value, soil texture, soil organic matter content, soil cation exchange capacity, land use history, climate data and fertilization management data.

4. The method for constructing a non-grain soil acidification prediction model according to claim 1, characterized in that: In S2, the collected soil sample data is preprocessed, specifically including the following operations: For missing data on soil organic matter content, estimates are made based on land use types and similar soil conditions in the surrounding area, or data interpolation methods are used to supplement them. For abnormal values, judgments and corrections are made by comparing them with the distribution of data on the same type of land to obtain normal values.

5. The method for constructing a non-grain soil acidification prediction model according to claim 3, characterized in that: The S3 specifically includes the following steps: The Pearson correlation coefficient between soil pH and other factors was calculated to screen the factor characteristics, the categorical variables were converted into numerical variables using one-hot encoding, and the continuous variables were standardized.

6. The method for constructing a non-grain soil acidification prediction model according to claim 1, characterized in that: In S4, when the random forest algorithm is used to construct the model, the random forest is composed of multiple decision trees, the number of decision trees is determined, samples are randomly selected with replacement from the training data set to construct the decision tree, a part of the factor features are randomly selected from all factor features when the node is split, the best split feature is selected according to the information gain, information gain ratio or Gini index, the decision tree grows until the stopping condition is met, and the prediction results of multiple decision trees are integrated through the idea of ​​ensemble learning.

7. The method for constructing a non-grain soil acidification prediction model according to claim 1, characterized in that: In S5, the data set is divided into 70% for training set and 30% for testing set, 10-fold cross validation is adopted, the mean square error function is used as the loss function for regression problem, the cross entropy function is used as the loss function for classification problem, and the model parameters are updated by stochastic gradient descent optimization algorithm.

Images