Method for predicting acidification improvement effect based on key attributes of conditioner

Abstract

The invention discloses a method for predicting an acidification improvement effect based on key attributes of a conditioner, and belongs to the technical field of acid soil improvement. The method comprises the following steps: establishing a preliminary database of the conditioner for improving the acid soil; after data in the preliminary database are subjected to normalization processing, the data in the preliminary database are read in the R language environment, data cleaning is carried out, abnormal values are deleted, and a database for analysis is obtained; performing importance screening on the initial explanatory variable indexes in the analysis database in the random forest model, determining a final explanatory variable, and performing random forest model construction; performing optimization parameter adjustment on mtry in the random forest model; the ntree in the random forest uses a fixed value; evaluating the precision of the optimized random forest model to obtain an optimal random forest model; the index parameters of the final explanatory variables are input into the optimal random forest model, and the effect of predicting soil acidification improved by the conditioner can be achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of acidic soil improvement, and more specifically to a method for predicting acidification improvement effects based on key properties of a conditioner. Background Art

[0002] Appropriate soil pH is an important foundation for sustainable agricultural production. Soil acidification can lead to reduced nutrient effectiveness, increased exchangeable aluminum content, increased risk of heavy metal pollution such as lead and cadmium, and reduced soil microbial and enzyme activity, which in turn affects crop growth and development, resulting in reduced or even total crop failure.

[0003] Applying conditioners is one of the key measures to control soil acidification. Inorganic conditioners mainly based on lime can reduce soil acidity, supply basic ions such as calcium, magnesium, and potassium, and reduce the toxicity of soil aluminum and manganese to crops, thereby improving crop growth. Organic conditioners such as agricultural waste, organic fertilizers, and biochar can not only neutralize soil acidity, but also have a good effect on improving the soil's ability to resist acidification.

[0004] In recent years, my country's soil conditioner market has developed rapidly. According to statistics, the annual output of acidic soil conditioners reaches 1 million tons. However, due to the wide variety of materials, the effective ingredients and functional effects of conditioners vary greatly, and the effects of different conditioners on improving acidic soils are difficult to predict. Some indicators, such as the influence of trace elements such as exchangeable calcium and magnesium, lack method support. At present, the recommended amount of lime to improve soil acidification is relatively common, such as the calcium hydroxide titration method, which can relatively accurately guide the application of lime. However, the improvement effects of common materials such as clay minerals, organic materials, and industrial waste are still difficult to quantify, and there is still a lack of acidification improvement prediction methods that can systematically consider different types of materials. Summary of the invention

[0005] In view of the above problems, the present invention provides a method for predicting the effect of acidification improvement based on the key properties of the conditioner. The present invention solves the problem that the improvement effect is difficult to predict due to the variety of conditioner material types and complex properties, and the conditioner dosage recommendation is inaccurate. The method of the present invention can effectively predict the effect of the conditioner on the improvement of acidified soil.

[0006] The object of the present invention is to provide a method for predicting the acidification improvement effect based on the key attributes of a conditioning agent, comprising the following steps:

[0007] Obtain physical and chemical data on the improvement of acidic soil by conditioners based on cultivation experiments, and establish a preliminary database on the improvement of acidic soil by conditioners.

[0008] After normalizing the data in the preliminary database, the data in the preliminary database were read in the R language environment and data cleaning was performed to delete outliers to obtain a database for analysis.

[0009] In the random forest, the initial explanatory variable indicators in the analysis database are screened for importance, the final explanatory variables are determined, and the random forest model is constructed.

[0010] The grid parameter adjustment method is used to optimize the optimal number of variables for binary trees in the specified nodes in the random forest model to obtain an optimized random forest model; wherein, the optimal number of decision trees contained in the specified random forest in the random forest uses a fixed value; wherein, the optimal number of variables for binary trees in the specified node is recorded as mtry, and the optimal number of decision trees contained in the specified random forest is recorded as ntree.

[0011] The accuracy of the optimized random forest model was evaluated using the determination coefficient and root mean square error indicators to determine the effect of the random forest model and obtain the best random forest model; the root mean square error is recorded as RMSE and the determination coefficient is recorded as R 2 .

[0012] The indicator parameters of the best random forest model are input into the best random forest model to predict the effect of conditioners on improving soil acidification.

[0013] In a preferred embodiment of the present invention, establishing a preliminary database of conditioners for improving acidic soils includes:

[0014] Based on the cultivation test, data on the improvement of acidic soil by conditioners were collected, and the properties of the conditioners were used as the initial explanatory variables and the changes in soil properties were used as the response variables to establish a preliminary database of the improvement of acidic soil by conditioners. The corresponding data were obtained by analyzing the properties of the conditioners and the soil before and after the test.

[0015] In a preferred embodiment of the present invention, the properties of the conditioning agent are application amount, pH, effective calcium, effective magnesium, total base amount, effective phosphorus, cation exchange capacity and specific surface area, wherein the specific surface area includes micropore specific surface area and mesopore specific surface area.

[0016] The soil properties are pH, exchangeable acid, exchangeable aluminum, exchangeable calcium, exchangeable magnesium, exchangeable base ions, cation exchange capacity, and base saturation. Cation exchange capacity is recorded as CEC.

[0017] In a preferred embodiment of the present invention, in the culture test, the conditioning agent is one or more of lime substances, industrial waste, clay minerals and organic matter. Further, lime substances include quicklime, slaked lime, limestone or dolomite, etc.; industrial waste includes flotation phosphate tailings powder, yellow phosphorus slag, etc.; clay minerals include bentonite, attapulgite, sepiolite or diatomaceous earth, etc.; organic materials include humic acid, straw biochar, rice husk biochar, pig manure or chicken manure, etc.

[0018] In a preferred embodiment of the present invention, the method of data cleaning is:

[0019] Check the missing values ​​in the database and fill in the missing values ​​in the variable indicators: use clustering methods to detect outliers, and complete the data cleaning after deleting outliers;

[0020] The missing values ​​are filled using one of the following methods: mean value, median simple filling, K-nearest neighbor algorithm, random forest filling missing values ​​and multiple interpolation method.

[0021] It should be noted that since the data in the present invention are derived from culture experiments and there are no missing values, there is no need to fill in missing values.

[0022] In a preferred embodiment of the present invention, the method for screening the importance of the initial explanatory variable index is:

[0023] The importance of explanatory variables is judged by using two indicator values: the increase in mean square error and the improvement in node purity. The indicators of explanatory variables after preliminary screening are obtained through screening. The larger the indicator value, the greater the importance of the explanatory variable.

[0024] The four-fold cross-validation method was used to re-screen the initially screened explanatory variable indicators to obtain the final explanatory variable indicators;

[0025] The final explanatory variables were application rate, pH, available calcium, available magnesium, total base, cation exchange capacity and specific surface area.

[0026] In a preferred embodiment of the present invention, the optimization parameter adjustment method for the optimal number of variables for the binary tree in a specified node in the random forest model is:

[0027] 70% of the data in the analysis database were selected as training samples, and the remaining 30% of the data were selected as test samples; the mtry in the random forest was screened and optimized using the binary classification and multi-classification of the grid parameter adjustment method, and the optimal parameter mtry was determined to be 3;

[0028] The fixed value of ntree in random forest is 500.

[0029] In a preferred embodiment of the present invention, the predictable soil properties include pH, exchangeable acid, exchangeable aluminum, exchangeable calcium and exchangeable magnesium.

[0030] In a preferred embodiment of the present invention, the method of predicting the effect of the conditioner on improving soil acidification includes:

[0031] In the R language environment, the final explanatory variable index is input to obtain the response variable; combined with the initial physical and chemical characteristics of the soil, the effect of soil acidification improvement is obtained.

[0032] That is, by inputting the final explanatory variable index in the R language environment, the change in soil properties after the application of the conditioner can be predicted, and the improvement effect can be clearly understood by combining the initial soil properties.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] The present invention is a method for predicting the improvement effect of acidic soil based on the key attributes of the conditioner. The method is based on machine learning technology and can quickly and accurately predict the application effect of common conditioners and effectively prevent overfitting. The method can be used to guide material selection and directional modification in the conditioner development process, as well as quantitative application of soil conditioners in farmland.

[0035] The application of conditioners is the key to improving soil acidification. In view of the fact that there are many types of soils and conditioners, with large variations, and there is a lack of quantitative prediction and evaluation methods for the improvement effect of soil acidification, the present invention establishes a quantitative prediction method through random forests, covering common material types, which can well support the actual needs of soil acidification improvement product development and field application.

[0036] Based on the prediction method provided by the present invention, the changes in soil multi-attributes can be predicted based on the amount and properties of the conditioner, thereby quantifying its improvement effect and guiding field application. In addition, this method is applicable to common mineral, organic, and organic-inorganic conditioners, and the predicted soil indicators are also expanded from a single soil pH value to important indicators such as exchangeable aluminum and secondary elements, which has important guiding significance for acidic soil improvement. In actual field test applications, influencing factors such as field management and crop planting can be introduced for further verification to improve the accuracy of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a diagram for screening the effects of conditioner indicators on soil properties, where (a) is soil pH, (b) is soil exchangeable acid, (c) is soil exchangeable aluminum, (d) is soil exchangeable calcium, (e) is soil exchangeable magnesium, and (f) is soil CEC.

[0038] Figure 2 This is a correlation analysis diagram of the predicted soil pH value and the measured soil pH value in the potted plant test in Example 2.

[0039] Figure 3 This is a correlation analysis diagram of the predicted values ​​and measured values ​​of soil properties in the potted plant test in Example 2, wherein (a) is the exchangeable acid (a) of the soil, (b) is the exchangeable aluminum of the soil, (c) is the exchangeable calcium of the soil, and (d) is the exchangeable magnesium of the soil.

[0040] Figure 4 This is a correlation analysis diagram of the predicted soil pH value and the measured pH value in the field test in Example 3.

[0041] Figure 5 This is a correlation analysis diagram of the predicted values ​​and measured values ​​of soil properties in the field test in Example 3, wherein (a) is the exchangeable acid of the soil, (b) is the exchangeable aluminum of the soil, (c) is the exchangeable calcium of the soil, and (d) is the exchangeable magnesium of the soil. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] As described in the background technology, at present, the recommended amount of lime to improve soil acidification is relatively common, such as the calcium hydroxide titration method, which can relatively accurately guide the application of lime. However, the improvement effect of common materials such as clay minerals, organic materials, and industrial waste is still difficult to quantify, and there is still a lack of acidification improvement prediction methods that can systematically consider different types of materials. The method of predicting the soil acidification improvement effect based on the characteristics of the conditioner can effectively solve the problem of complicated improvement materials and difficult prediction of improvement effects, and provide method support for the development and field application of acidic soil conditioners.

[0044] Based on this, the present invention constructs a database based on the data of the cultivation experiment, and obtains more comprehensive and complete data through laboratory testing to make up for the problem of missing some indicators in the current meta data. The database is constructed to cover the experimental effects of conditioners for the improvement of common organic and inorganic materials. And use random forests, a method that directly learns complex functions from data by training a large number of parameters, to produce accurate predictions when the underlying mechanism is not fully understood or parameterized, to achieve accurate predictions of the acidification improvement effect based on the properties of the conditioner, and to apply it to field test guidance.

[0045] Example 1

[0046] This embodiment provides a method for predicting the acidification improvement effect based on the properties of the conditioning agent established by a random forest model, and the specific steps are as follows:

[0047] Step 1: Conduct cultivation experiments on soils of different acidity to clarify the improvement effect and establish a preliminary database of conditioners to improve acidic soils.

[0048] The specific cultivation test plan is: select acidic red soil with different acidity, pass through a 2mm sieve, design different types and dosages of conditioners, adjust the moisture content by 20%, which is about 60% of the field water holding capacity. If there is no complete infiltration, the moisture content can be adjusted to 30%, and cultivate at a constant temperature indoors. Each treatment is repeated 3 times, and the culture samples are weighed every 3 days. Water is supplemented according to the loss, and the cultivation time is 30 days. A total of 267 culture samples were tested.

[0049] The types of materials used as conditioners in the cultivation experiment include lime, limestone, dolomite and other lime-based materials; industrial wastes such as collophane flotation phosphate tailings and yellow phosphorus slag; clay minerals such as diatomaceous earth, attapulgite, sepiolite, and bentonite; organic materials such as mineral-derived humic acid, rice husk biochar, straw biochar, pig manure, and chicken manure. The conditioner is mixed with one or more of these materials, and the application rate of the conditioner is: 160kg / mu, 320kg / mu, 800kg / mu, 1600kg / mu or part of them. The experimental soils are two acidic red soils with pH values ​​of 4.2 and 5.0.

[0050] The relevant properties of the conditioner include: application amount, pH, effective calcium, effective magnesium, total base, effective phosphorus, CEC and specific surface area, among which the specific surface area includes micropore specific surface area and mesopore specific surface area.

[0051] The soil property indicators are pH, exchangeable acid, exchangeable aluminum, exchangeable calcium, exchangeable magnesium, exchangeable base, CEC and base saturation. After removing the control group, a total of 248 acidification improvement data were collected. It should be noted that the control group refers to the acidification improvement test in which no conditioner is used to test the changes in the soil. Since there is no improvement effect in the database, the control group is removed.

[0052] Step 2: Normalize the soil data in the preliminary database. Use the property of the conditioner as the initial explanatory variable and the change in the soil property as the response variable and enter it into the preliminary database.

[0053] Step 3: Read the preliminary database in the R language environment and clean the data in the preliminary database, use the clustering method to detect outliers, and complete the data cleaning after deleting the outliers to form a database for analysis.

[0054] Step 4: Screening important feature indicators: In the random forest, the initial explanatory variable indicators in the analysis database are ranked in importance, and the importance indicators of the initial explanatory variables are judged by the two indicator values ​​of the increase in mean square error and the improvement in node purity. The larger the indicator value, the greater the importance of the explanatory variable; the explanatory variable indicators after preliminary screening are obtained through screening. It is understandable that the selection of explanatory variables needs to make the model explanatory degree above a certain range, because the data for this evaluation is not shown in the results of running the model. However, there is R in the final model data. 2 The correlation between the predicted value and the observed value can be evaluated by using RMSE to evaluate the accuracy of the model.

[0055] Then, the explanatory variable indicators for constructing the random forest model were determined by the four-fold cross-validation method. It should be noted that when the four-fold cross-validation method re-screens the explanatory variable indicators that have been initially screened, the screening criterion is the comprehensive explanatory power of the selected explanatory variable indicators.

[0056] Finally, the final explanatory variable indicators were selected as the final explanatory variables of the model: application amount, pH, effective calcium, effective magnesium, total base, CEC and specific surface area. Figure 1 shown.

[0057] Step 5: Optimize the parameters of the random forest model: Select 70% of the data in the analysis database as training samples, and the remaining 30% of the data as test samples. Use the random forest grid parameter adjustment method to optimize the parameter mtry. When mtry is 3, the error rate is the lowest, so it is used as the modeling parameter. The default value of 500 is used for ntree. The random forest model is constructed with the determined optimal parameters, which is recorded as the best random forest model.

[0058] Step 6: Compare RMSE and R 2 The accuracy of this method is evaluated. When the predicted value of soil property change is closer to the true value, the RMSE is closer to 0, and R 2 The closer it is to 1, the better the model effect is;

[0059] RMSE and R 2 The calculation formulas are shown in formula (1) and formula (2) respectively.

[0060]

[0061] Where: c i represents the measured value of the change in soil properties after treatment with the ith conditioner, such as the change in pH value; c ii represents the predicted value of soil property changes after treatment with the ith conditioner; c mIt represents the average value of the measured value of soil property change, such as soil pH, n represents the total number of conditioners, and i represents the i-th conditioner.

[0062] Step 7: Predict the acidification improvement effect of the conditioner based on the best random forest model. Input the index parameters of the final explanatory variables into the best random forest model. The index parameters of the final explanatory variables are the application amount of the conditioner, pH, effective calcium, effective magnesium, total base, CEC, and specific surface area. Run the model to predict changes in soil properties.

[0063] Specifically, the soil improvement effect can be predicted by combining the initial soil properties, such as soil pH, exchangeable acids and exchangeable bases.

[0064] Example 2

[0065] The embodiment of the present invention provides a method for predicting the acidification improvement effect based on the key attributes of the conditioner.

[0066] Based on the random forest model, a method was established to predict the improvement effects of soil pH, exchangeable acid, exchangeable calcium and exchangeable magnesium using the key attributes of conditioners.

[0067] A random forest prediction model was established based on the steps of Example 1. To further illustrate the accuracy of the prediction model, the results of a soil acidification improvement potted plant experiment were used for verification. The soil used in the experiment was typical acidic red soil in the south, and the crop planted was Chinese cabbage. The following six groups of conditioning agents are selected: (1) phosphate tailings from collophanite flotation, (2) 80% phosphate tailings from collophanite flotation and 20% yellow phosphorus slag, (3) 80% phosphate tailings from collophanite flotation and 20% sepiolite, (4) 80% phosphate tailings from collophanite flotation and 20% humic acid, (5) 60% phosphate tailings from collophanite flotation, 20% sepiolite and 20% humic acid, (6) 60% phosphate tailings from collophanite flotation, 15% yellow phosphorus slag, 15% sepiolite and 10% humic acid. It should be noted that the percentages here refer to the percentages by mass. Taking 80% tailings and 20% yellow phosphorus slag as an example, in the applied conditioning agent, the tailings account for 80% of the total mass of the conditioning agent, and the yellow phosphorus slag accounts for 20% of the total mass of the conditioning agent.

[0068] The total application rate of the conditioner was 320kg / mu and 800kg / mu, respectively, and 6 different conditioner treatments were designed. The final explanatory variables and soil indicators required for the optimal random forest model were determined. The final explanatory variables were application rate, pH, available calcium, available magnesium, total base, CEC, specific surface area, and the soil properties before and after the experiment were pH, exchangeable acid, exchangeable aluminum, exchangeable calcium, exchangeable magnesium, etc.

[0069] The relationship between the predicted value of the best random forest model and the soil pH value measured by the experiment is as follows: Figure 2As shown in the figure, the prediction results of acidification improvement based on the key characteristics of the conditioner are relatively ideal, and the predicted values ​​are linearly correlated with the measured values. 2 The predicted results for exchangeable acid and exchangeable aluminum are as follows: Figure 3 A and Figure 3 As shown in b, its determination coefficient R 2 They are 0.83 and 0.77 respectively; the prediction results of exchangeable calcium and exchangeable magnesium are as follows Figure 3 C and Figure 3 As shown in d, the determination coefficients are 0.49 and 0.74 respectively. It can be seen that the prediction model has a high degree of explanation and can better predict the changes in soil properties after the application of conditioners.

[0070] Example 3

[0071] The method according to the present invention simulates the change of soil properties after applying soil conditioners in the field.

[0072] The verification data were taken from field test data from 12 test sites in Jinning, Yunnan, Pinghe, Fujian, Qiyang, Hunan, etc. The material types of conditioners used included lime-based substances, industrial waste such as phosphate tailings and yellow phosphorus slag, and organic substances such as biochar and manure.

[0073] Among them, the test soil in Jinning District, Yunnan Province is typical yellow soil and red soil, the type of conditioning agent is a complex of phosphorus tailings and yellow phosphorus slag, and the crops planted include zucchini, corn, tobacco, pepper, and lettuce. Among them, the total application amount of conditioning agent in zucchini, corn, tobacco, and pepper is 600kg / mu, and the addition amount of yellow phosphorus slag accounts for 2.5%, 5% or 10% of the total mass of conditioning agent. The application amount of conditioning agent in lettuce is 200kg / mu, 400kg / mu, 600kg / mu and 800kg / mu respectively, and the addition amount of yellow phosphorus slag accounts for 5% of the total mass of conditioning agent.

[0074] The Pinghe experiment in Fujian Province was a typical acidic red soil, and the crop planted was honey pomelo. The following four groups of conditioning agents were selected: (1) slaked lime, with an application rate of 212 kg / mu; (2) organic fertilizer, oyster shell powder and phosphogypsum, of which the application rate of organic fertilizer was 133 kg / mu, the application rate of oyster shell powder was 101 kg / mu, and the application rate of phosphogypsum was 41 kg / mu; (3) organic fertilizer and slaked lime, of which the application rate of organic fertilizer was 133 kg / mu, and the application rate of slaked lime was 212 kg / mu; (4) organic fertilizer and magnesium oxide, of which the application rate of organic fertilizer was 212 kg / mu, and the application rate of magnesium oxide was 3 kg / mu.

[0075] The experimental site in Qiyang, Hunan Province is acidified red soil, and the crop planted is corn. The following two groups of conditioning agents were selected: (1) quicklime and calcium carbonate, where the application rate of calcium carbonate is 333 kg / mu and the application rate of quicklime is 170 kg / mu; (2) quicklime and calcium carbonate, where the application rate of calcium carbonate is 666 kg / mu and the application rate of quicklime is 170 kg / mu.

[0076] The final explanatory variables collected were: application rate, pH value, available calcium, available magnesium, total base, CEC, and specific surface area.

[0077] The relationship between the predicted results and the actual results of soil properties is as follows: Figure 4 and Figure 5 shown. Figure 4 In the study, the predicted and measured values ​​of soil pH were linearly related, R 2 The results of the field test data based on multiple sites and multiple crops show that this method can well predict the effect of the application of conditioners on soil pH. The prediction results of soil exchangeable acid, exchangeable aluminum, exchangeable calcium and exchangeable magnesium are as follows: Figure 5 As shown, Figure 5 A and Figure 5 b shows that the determination coefficient R of soil exchangeable acid and exchangeable aluminum 2 are 0.71 and 0.46 respectively. Figure 5 C and Figure 5 The determination coefficients R of exchangeable calcium and exchangeable magnesium are shown in d. 2 The predicted values ​​and measured values ​​are 0.81 and 0.78 respectively, and the correlation between the predicted value and the measured value is good. In summary, the prediction method of the present invention can better reflect the acidification improvement effect of common conditioners, thereby guiding the use of conditioners.

[0078] Compared with Chinese patent CN117275604A, although it provides a soil pH prediction method based on a random forest model, the only improvement material it targets is lime, which is difficult to meet the current variety of conditioning agent material types; and its database is established based on mate data, resulting in insufficient types of indicators in some areas, and the prediction of soil properties can only be limited to soil pH. When predicting the acidic improvement effect, the present invention takes into account the problem of incomplete or lacking corresponding indicators when mate data is statistically analyzed, resulting in the limitation of prediction indicators, and collects relatively complete data indicators based on a large number of cultivation experiments, covering the current common types of conditioning agents, and then relying on the random forest method to establish a prediction model, in which the change in soil properties is used as a response indicator to improve the model accuracy, and finally achieves the purpose of being able to accurately predict the acidic improvement effect. On this basis, the method predicts the soil improvement effect using conditioning agent properties as explanatory variables alone, which is more conducive to guiding the field application of conditioning agents.

[0079] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0080] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for predicting the acidification improvement effect based on the key attributes of a conditioning agent, characterized in that: The following steps are involved: Obtain physical and chemical data on the improvement of acidic soil by conditioners based on cultivation experiments and establish a preliminary database on the improvement of acidic soil by conditioners; After normalizing the data in the preliminary database, the data in the preliminary database was read in the R language environment and data cleaning was performed to delete outliers to obtain a database for analysis; In the random forest, the initial explanatory variable indicators in the analysis database are screened for importance, the final explanatory variables are determined, and the random forest model is constructed; The grid parameter adjustment method is used to optimize the number of optimal variables used for the binary tree in the specified node in the random forest model to obtain an optimized random forest model; wherein the number of optimal decision trees contained in the specified random forest in the random forest model uses a fixed value; The accuracy of the optimized random forest model was evaluated using the coefficient of determination and root mean square error indicators to determine the effect of the random forest model and obtain the best random forest model; The indicator parameters of the final explanatory variables were input into the optimal random forest model to predict the effect of the conditioner on improving soil acidification.

2. A method for predicting acidification improvement effect based on key attributes of a conditioning agent according to claim 1, characterized in that: The initial database of conditioners for improving acidic soils includes: Based on the cultivation experiment, data on the improvement of acidic soil by conditioners were collected. The properties of the conditioners were used as the initial explanatory variables and the changes in soil properties were used as the response variables to establish a preliminary database for the improvement of acidic soil by conditioners.

3. A method for predicting acidification improvement effect based on key attributes of a conditioning agent according to claim 2, characterized in that: Conditioner attributes were application rate, pH, available calcium, available magnesium, total base, available phosphorus, cation exchange capacity, and specific surface area; The soil properties are pH, exchangeable acid, exchangeable aluminum, exchangeable calcium, exchangeable magnesium, exchangeable base ions, cation exchange capacity, and base saturation.

4. A method for predicting acidification improvement effect based on key attributes of a conditioning agent according to claim 2, characterized in that: In the culture test, the conditioning agent is one or more of lime substances, industrial wastes, clay minerals and organic matter.

5. A method for predicting acidification improvement effect based on key attributes of a conditioning agent according to claim 1, characterized in that: The data cleaning method is: Check the missing values ​​in the database and fill in the missing values ​​in the variable indicators: use clustering methods to detect outliers, and complete the data cleaning after deleting outliers; The missing values ​​are filled using one of the following methods: mean value, median simple filling, K-nearest neighbor algorithm, random forest filling missing values ​​and multiple interpolation method.

6. A method for predicting acidification improvement effect based on key attributes of a conditioning agent according to claim 3, characterized in that: The method for screening the importance of the initial explanatory variable indicators is: The importance of explanatory variables is judged by using two indicator values, namely, the increase of mean square error and the improvement of node purity. The explanatory variable indicators after preliminary screening are obtained through screening. The four-fold cross-validation method was used to re-screen the initially screened explanatory variable indicators to obtain the final explanatory variable indicators; The final explanatory variables were application rate, pH, available calcium, available magnesium, total base, cation exchange capacity and specific surface area.

7. A method for predicting acidification improvement effect based on key attributes of a conditioning agent according to claim 1, characterized in that: The optimization method for the optimal number of variables used for binary trees in a specified node in the random forest model is: 70% of the data in the analysis database were selected as training samples, and the remaining 30% of the data were selected as test samples. The optimal number of variables for binary trees in random forests was screened and optimized using the binary classification and multi-classification of the grid parameter adjustment method, and the optimal number of variables for binary trees was determined to be 3. The random forest specifies a fixed value of 500 for the optimal number of decision trees contained in the random forest.

8. A method for predicting acidification improvement effect based on key attributes of a conditioning agent according to claim 6, characterized in that: The predictable soil properties are pH, exchangeable acid, exchangeable aluminum, exchangeable calcium, and exchangeable magnesium.

9. A method for predicting acidification improvement effect based on key attributes of a conditioning agent according to claim 1, characterized in that: The effects of predicting the improvement of soil acidification by conditioners include: In the R language environment, the final explanatory variable index is input to obtain the response variable; combined with the initial physical and chemical characteristics of the soil, the effect of soil acidification improvement is obtained.

Images