Precise acid reduction machine learning prediction method and system for acid cultivated land soil

Abstract

The invention discloses an accurate acid reduction machine learning prediction method and system for acid cultivated land soil, and the method comprises the steps: S1, collecting acid soil, and carrying out the classification according to the acidification degree of soil pH; s2, measuring physical and chemical indexes of the collected soil, and screening key physical and chemical indexes of soil acidification; s3, carrying out a soil culture experiment of the quicklime application amount; s4, preprocessing the experimental data, and dividing the experimental data into a training set and a test set; s5, training different machine learning models by using the training set, performing multiple prediction verification by using a five-fold cross validation and external optimization verification method, and selecting the model with the optimal parameter combination after comparison; and S6, learning and extracting important information such as soil acidification key physicochemical indexes and the like by adopting the trained machine learning model, performing result prediction, and outputting a predicted value of the required quicklime amount. According to the invention, the constructed machine learning model is utilized to predict the application amount of the quick lime required by the acidic soil to reach the target pH, so that accurate deacidification is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of soil acid reduction, and particularly relates to a machine learning prediction method and system for precise acid reduction of acidic cultivated land soil. Background Technique

[0002] Acidic soils are widely distributed in China, especially in Guangdong, Guangxi, Fujian and other places. Soil acidification can lead to soil nutrient deficiency and crop yield reduction, and also increase the activity of soil heavy metals, making them more easily absorbed and accumulated by crops, thus threatening human health.

[0003] Lime materials have become commonly used acidified soil amendments due to their economy and effectiveness. Lime materials mainly neutralize acidic substances in the soil by providing calcium oxide, thereby increasing the pH value of the soil and slowing down the acidification degree. Among them, quicklime is widely used because of its highest alkalinity and fastest acid reduction effect. In traditional acidified soil improvement, the application rate of quicklime often cannot be accurately calculated, resulting in over-application or under-application, thus causing waste of manpower and materials.

[0004] With the rapid development of artificial intelligence technology, the application of machine learning in the field of environmental science has gradually increased. For example, machine learning can analyze a large amount of environmental data, identify environmental change trends and patterns, predict extreme weather events, improve the accuracy of disaster warnings, or use machine learning algorithms such as neural networks and random forests to predict the future trends of climate variables and identify complex weather patterns. By collecting and analyzing a large amount of soil physical and chemical indexes, quicklime application rates and application effects and other data, and constructing a machine learning model, it is expected to achieve precise prediction of the quicklime application rate in the process of acidic soil improvement, so as to maximize the improvement effect of quicklime. Summary of the Invention

[0005] The main purpose of the present invention is to overcome the deficiencies of the prior art and propose a machine learning prediction method and system for precise acid reduction of acidic cultivated land soil.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions:

[0007] A machine learning prediction method for precise acid reduction of acidic cultivated land soil, comprising the following steps:

[0008] S1. Collect acidic soil and classify it according to the acidification degree of the soil pH;

[0009] S2. Measure the physical and chemical indexes of the collected soil and screen the key physical and chemical indexes of soil acidification;

[0010] S3. Use the collected soil as the test soil and quicklime as the test calcareous material to conduct a soil culture experiment on the application rate of quicklime, collect experimental data, and establish the relationship between the application rate of quicklime and the increase in soil pH;

[0011] S4. Preprocess the collected experimental data and divide the preprocessed data into a training set and a test set at a ratio of 8:2;

[0012] S5. Use the training set to train different machine learning models, establish the relationship among the application rate of quicklime, the key physical and chemical indexes of soil acidification, and the increase in soil pH, use the five-fold cross-validation and external optimization validation methods to conduct multiple prediction validations, screen out the machine learning model with the best parameter combination after comparison, and adjust the parameters of the model;

[0013] S6. Use the trained machine learning model to learn and extract the key physical and chemical indexes of soil acidification, conduct result prediction, and output the application rate of quicklime required for acidic soil to reach the target pH.

[0014] The present invention also includes a machine learning prediction system for precise acid reduction of acidic cultivated land soil. The system adopts the machine learning prediction method for precise acid reduction of acidic cultivated land soil provided by the present invention. The system includes an acidic soil collection module, a soil physical and chemical index measurement module, a quicklime application rate cultivation experiment module, an experimental data preprocessing module, a machine learning model training module, and a prediction module;

[0015] The acidic soil collection module is used to collect acidic soil, classify the degree of soil acidification according to the standards of pH < 4.5, 4.5 ≤ pH < 5.0, 5.0 ≤ pH < 5.5, and 5.5 ≤ pH < 6.5, and select the soil with pH < 5.5 as the test soil;

[0016] The soil physical and chemical index measurement module is used to measure the physical and chemical indexes of the collected soil and screen out the key physical and chemical indexes of soil acidification;

[0017] The quicklime application rate cultivation experiment module is used to conduct a soil cultivation experiment on the application rate of quicklime, collect experimental data, and establish the relationship between the application rate of quicklime and the increase in soil pH;

[0018] The experimental data preprocessing module is used to preprocess the collected experimental data and divide the preprocessed data into a training set and a test set;

[0019] The machine learning model training module uses the training set to train different machine learning models, uses the five-fold cross-validation and external optimization validation methods to conduct multiple prediction validations, screens out the machine learning model with the best parameter combination after comparison, and adjusts the parameters of the model to improve the prediction accuracy;

[0020] The prediction module uses a trained machine learning model to learn and extract the key physical and chemical indicators of soil acidification, make result predictions, and output the amount of quicklime application required for acidic soil to reach the target pH.

[0021] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0022] 1. Based on machine learning, the present invention constructs a machine learning model, which can accurately predict the amount of lime application for acid reduction in acidic soil, thereby achieving precise acid reduction, maximizing the improvement effect of lime, and realizing precise control of acidic soil improvement through the assistance of big data and machine learning, improving soil health, and ensuring the sustainable development of agricultural production. Description of the Drawings

[0023] Figure 1 is the flowchart of Embodiment 1 of the present invention;

[0024] Figure 2 is the fitting diagram of the actual value and predicted value of the combined indicators of the trained random forest model in Embodiment 1;

[0025] Figure 3 is the fitting diagram of the actual value and predicted value of the split indicators of the trained random forest model in Embodiment 1;

[0026] Figure 4 is the laboratory verification diagram of the combined indicators of the trained random forest model in Embodiment 1;

[0027] Figure 5 is the laboratory verification diagram of the split indicators of the trained random forest model in Embodiment 1;

[0028] Figure 6 is the field verification diagram of the combined indicators of the trained random forest model in Embodiment 1;

[0029] Figure 7 is the field verification diagram of the split indicators of the trained random forest model in Embodiment 1;

[0030] Figure 8 is the comparison diagram of the predicted application amounts of the combined indicators of the trained random forest model in Embodiment 1 and domestic and foreign software;

[0031] Figure 9 is the comparison diagram of the predicted application amounts of the split indicators of the trained random forest model in Embodiment 1 and domestic and foreign software;

[0032] Figure 10 is the model literature verification diagram of the combined indicators of the trained random forest model in Embodiment 1;

[0033] Figure 11It is a model literature verification diagram of the split index of the random forest model completed in training in Example 1. Detailed implementation manners

[0034] The present invention will be further described in detail below in conjunction with examples and the accompanying drawings, but the implementation manners of the present invention are not limited thereto.

[0035] Example 1

[0036] As Figure 1 shown, the present invention, an accurate acid reduction machine learning prediction method for acidic cultivated land soil, includes the following steps:

[0037] S1. Collect acidic soil and classify it according to the acidification degree of soil pH; specifically:

[0038] Collect acidic soil from multiple provinces in China, classify the acidification degree of the soil according to the standards of pH < 4.5, 4.5 ≤ pH < 5.0, 5.0 ≤ pH < 5.5, 5.5 ≤ pH < 6.5, and select the soil with pH < 5.5 as the test soil.

[0039] S2. Measure the physical and chemical indexes of the collected soil and screen the key physical and chemical indexes of soil acidification;

[0040] Among them, the key physical and chemical indexes of soil acidification are specifically:

[0041] Soil active acid and potential acid indexes: soil pH, exchangeable aluminum ( ex [Al 3+ ), exchangeable hydrogen ( ex [H + ), total exchangeable acid ([H + ex ); among them, the total exchangeable acid is the sum of exchangeable aluminum and exchangeable hydrogen;

[0042] Indexes of soil acid buffering capacity: soil organic matter (OM);

[0043] Soil physical indexes: soil clay (CL).

[0044] S3. Use the collected soil as the test soil and quicklime as the test calcareous material, conduct a soil culture experiment on the application amount of quicklime, collect experimental data and establish the relationship between the application amount of quicklime and the increase in soil pH;

[0045] Among them, the soil culture experiment on the application amount of quicklime is specifically:

[0046] ​Weigh 7 g of acidic soil, add different application rates of quicklime to the soil, mix evenly, incubate in the dark at 25 °C with 70% saturated water holding capacity for 24 h, add carbon dioxide-free water to the soil to make the soil-water ratio 1:2.5, stir for 5 min and let stand for 1 h, and measure the pH of the supernatant.

[0047] S4. Preprocess the data of the quicklime application rate (Q), key physical and chemical indexes of soil acidification ( ex [H + , ex [Al 3+ , [H + ex , pH, OM, CL), and the soil pH increase amount (ΔpH). Divide the preprocessed data into a training set and a test set at a ratio of 8:2;

[0048] Among them, the preprocessing is specifically as follows:

[0049] Based on the Python environment, use the Numpy and Pandas libraries in it to preprocess the collected experimental data, including:

[0050] Normalize the unit of the parameters of the key physical and chemical indexes of soil acidification data collected to ensure data consistency;

[0051] Perform multiple imputations on the missing values of the data to maintain data integrity;

[0052] After normalizing the parameter units and missing values, perform standardization processing on the data, specifically:

[0053] Subtract the mean value of each column of feature parameters from the value of the feature parameter, and then divide by the standard deviation of the feature parameter to obtain a set of new feature parameter values.

[0054] S5. Use the training set to train different machine learning models to establish the relationship among the quicklime application rate (Q), key physical and chemical indexes of soil acidification ( ex [H + , ex [Al 3+ , [H + ex , pH, OM, CL), and the soil pH increase amount (ΔpH). Use different parameter values to perform multiple prediction validations through five-fold cross-validation, compare and select the machine learning model with the best parameter combination, and tune the parameters of the model, specifically:

[0055] ​​In Python, model training and evaluation are carried out through the scikit-learn library. The training set is input into the constructed machine learning model for training. The machine learning models constructed in this embodiment include K-Nearest Neighbors (KNN), Random Forest, XGBoost (Extreme Gradient Boosting), and Gradient Boosting Trees.

[0056] Input the validation set, and successively use different machine learning model parameter values through five-fold cross-validation and external optimization verification methods; calculate the root mean square error (RMSE) on the training set to evaluate the performance of the model on the training set; calculate the coefficient of determination R of the model on the test set 2 ; In this embodiment, through the coefficient of determination R 2 and the root mean square error RMSE comparison, it is determined that the random forest model is the best machine learning model.

[0057] For the selected best machine learning model, through five-fold cross-validation, and adjust the model parameters to continuously improve the model prediction accuracy.

[0058] In this embodiment, it also includes optimizing the random forest model: first, find the best cross-validation score by looping through different random_state values, and loop through different numbers of trees to find the best n_estimators. Finally, fix the best random_state and n_estimators, and loop through different maximum depth values to optimize the model.

[0059] Model interpretation: Use the SHAP (SHapley Additive exPlanations) library to interpret the trained random forest (RFC) model for in-depth feature analysis.

[0060] S6. Use the trained machine learning model to learn and extract important information such as key physical and chemical indicators of soil acidification, make result predictions, and output the amount of quicklime application required for acidic soil to reach the target pH.

[0061] Such as Figure 2 and Figure 3 shown, Figure 2 (Combined indicators) and Figure 3 (Split indicators) are the fitting diagrams of the actual value (ΔpH / Q) and the predicted value (ΔpH / Q) of the trained random forest model in the embodiment; ΔpH refers to the difference between the target pH and the initial pH, and Q refers to the amount of quicklime application.

[0062] Next, the accuracy and effectiveness of the random forest model obtained in this embodiment in adjusting the pH value of acidic soil in the southeastern region to 6.3 are verified in the laboratory, including the following steps:

[0063] Step 1, sample collection; 27 samples were selected from acidic soils in southeastern regions such as Guangdong, Guangxi, Hainan, and Hunan to ensure that the initial pH values of all samples were lower than 5.5.

[0064] Step 2, model application; Set the target pH value to 6.3, and use the trained random forest model to predict the amount of quicklime (%) to be applied to each soil sample.

[0065] Step 3, accurately weigh 20 grams of dry soil samples; According to the prediction results of the random forest model, add the corresponding amount of quicklime to the soil and mix it evenly; Under the condition of 25°C, place the soil samples in a light-proof and sealed environment for incubation, while maintaining 70% saturated water holding capacity of the soil (adjusted by regular weighing); After each 2-hour equilibration culture, add carbon dioxide-free water to the soil to make the soil-water ratio reach 1:2.5. After stirring for 5 minutes and standing for 1 hour, measure the pH of the supernatant; Repeat the above operations until the soil pH value stably reaches 6.3, and record and calculate the total amount of quicklime actually used during the whole process.

[0066] Step 4, result comparison and verification; Compare and analyze the actual quicklime application amount obtained from the experiment with the theoretical application amount predicted by the random forest model. The results show that there is a high degree of consistency between the two, fully demonstrating the accuracy and effectiveness of the random forest model in predicting and calculating the quicklime application amount.

[0067] As Figure 4 and Figure 5 shown, Figure 4 (Combined indicators) and Figure 5 (Split indicators) are the laboratory verification diagrams of the random forest model.

[0068] Next, the prediction accuracy of the random forest model obtained in this embodiment in adjusting the soil pH value to the target level (6.3) is verified in the field environment, including the following steps:

[0069] Step 1, test field selection; Select a paddy field located in Nanxiong County, Shaoguan City, Guangdong Province as the site for this field experiment, and ensure that the initial pH value of the soil in this field block is lower than 5.5.

[0070] Step 2, initially apply quicklime; According to experience or preliminary estimation, apply a certain amount of quicklime to the test field to start the process of adjusting the soil pH.

[0071] Step 3: Model application and prediction; Subsequently, the trained random forest model is used to predict the specific amount of quicklime application (%) required to reach the target pH value based on the key physical and chemical indicators of soil acidification in the experimental field (such as soil exchangeable acid, organic matter content, etc.).

[0072] Step 4: Comparative verification; The amount of quicklime application predicted by the random forest model is compared with the known amount of quicklime applied in the field experiment. Through data comparison, the high consistency between the model prediction value and the actual application amount is confirmed, thus effectively verifying the accuracy of the model in predicting and calculating the amount of quicklime application and its effectiveness in actual agricultural production.

[0073] Such as Figure 6 and Figure 7 as shown Figure 6 (Combined indicators) and Figure 7 (Split indicators) are the field verification diagrams of the random forest model.

[0074] The following provides the comparative analysis process of the random forest model obtained in this embodiment and domestic and foreign software. By comparing the prediction results of the random forest model with other domestic and foreign calculation tools for quicklime application in acidic soils (including the model applet of Nanjing Agricultural University and the RothLime software of the UK Rothamsted Research), the accuracy and reliability of the random forest model are further verified; specifically, it includes the following steps:

[0075] Step 1: Sample collection; 17 representative samples are selected from acidic soils in southeastern regions such as Guangdong, Guangxi, Hainan, and Hunan to ensure that the initial pH values of all samples are lower than 5.5.

[0076] Step 2: Model application (model of this embodiment); Set the target pH value to 6.3, and use the random forest model to calculate the amount of quicklime required for each soil sample (%).

[0077] Step 3: Comparison with domestic models; Also with 6.3 as the target pH value, use the acidic soil model calculation applet developed by Nanjing Agricultural University to calculate the corresponding amount of quicklime application for comparison with the results of the random forest model.

[0078] Step 4: Comparison with international software; Use the RothLime software, also set the target pH to 6.3, input the relevant parameters of the soil sample, and calculate and obtain the amount of quicklime recommended by RothLime; Since RothLime takes into account the relatively high organic matter content in UK soils, the recommended application amount may be slightly higher than that in other regions.

[0079] Step 5, Result Comparison and Analysis: Comprehensively analyze the quicklime application rates calculated by the random forest model, the Nanjing Agricultural University model, and the RothLime software. Pay particular attention to the calculation results of the RothLime software. Considering factors such as rounding of the application rate and differences in soil organic matter content, understand and explain the possible reason for the slightly higher recommended application rate. Through comprehensive comparison, verify the accuracy of the random forest model in predicting the quicklime application rate and evaluate its differences and consistency with internationally renowned software.

[0080] As Figure 8 and Figure 9 shown, Figure 8 (Combined indicators) and Figure 9 (Split indicators) are the comparison charts of the application rate predictions of the random forest model and domestic and foreign software.

[0081] The following provides the process of using literature data to verify the accuracy of the random forest model obtained in this embodiment. By comparing the quicklime application rate calculated by the random forest model with the application rate actually used in the field in the published literature, verify the prediction accuracy of the random forest model in improving acidic soil. Specifically, it includes the following steps:

[0082] Step 1, Literature Data Collection and Screening: Systematically collect and analyze relevant field test literature on using calcareous substances (such as quicklime) to improve acidic soil, and screen out data samples containing the initial soil pH value, target pH value, type of calcareous substance used, and its specific application rate to ensure the accuracy and effectiveness of the comparative analysis.

[0083] Step 2, Model Application and Comparative Verification: For the screened literature data, use the random forest model obtained in this embodiment to calculate the theoretical application rate of quicklime based on the initial soil pH value and target pH value provided in the literature. Subsequently, compare the calculation results with the actual reported quicklime application rate in the literature, and evaluate the accuracy and applicability of the random forest model in predicting the quicklime application rate by comparing the differences between the two.

[0084] As Figure 10 and Figure 11 shown, Figure 10 (Combined indicators) and Figure 11 (Split indicators) are the model literature verification charts of the random forest model.

[0085] Example 2

[0086] Example 2 provides a machine learning prediction system for precisely reducing soil acidity in acidic arable land. The system adopts the machine learning prediction method for precisely reducing soil acidity in acidic arable land in Example 1 above. The system includes an acidic soil collection module, a soil physical and chemical index measurement module, a quicklime application rate cultivation experiment module, an experimental data preprocessing module, a machine learning model training module, and a prediction module;

[0087] The acidic soil collection module is used to collect acidic soil, classify the degree of soil acidification according to the standards of pH < 4.5, 4.5 ≤ pH < 5.0, 5.0 ≤ pH < 5.5, and 5.5 ≤ pH < 6.5, and select the soil with pH < 5.5 as the test soil;

[0088] The soil physical and chemical index measurement module is used to measure the physical and chemical indexes of the collected soil and screen the key physical and chemical indexes of soil acidification;

[0089] The quicklime application rate cultivation experiment module is used to conduct a soil cultivation experiment on the quicklime application rate, collect experimental data, and establish the relationship between the quicklime application rate and the increase in soil pH;

[0090] The experimental data preprocessing module is used to preprocess the collected experimental data and divide the preprocessed data into a training set and a test set;

[0091] The machine learning model training module uses the training set to train different machine learning models, conducts multiple prediction validations using the five-fold cross-validation and external optimization validation methods, selects the machine learning model with the best parameter combination after comparison, and tunes the parameters of the model to improve the prediction accuracy;

[0092] The prediction module uses the trained machine learning model to learn and extract the key physical and chemical indexes of soil acidification, make result predictions, and output the quicklime application rate required for the acidic soil to reach the target pH.

[0093] It should also be noted that in this specification, terms such as "including", "comprising", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements but also includes other elements not explicitly listed, or also includes elements inherent to such a process, method, article, or device. Without further limitations, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article, or device including the said element.

[0094] The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A machine learning prediction method for accurate acid reduction in acidic cultivated soil, characterized in that: The following steps are involved: S1. Collect acidic soil and classify it according to the acidification degree of soil pH; S2. Determine the physical and chemical indicators of the collected soil and screen the key physical and chemical indicators of soil acidification; S3. Using the collected soil as the test soil and quicklime as the calcareous material for the test, a soil culture experiment on the application amount of quicklime was conducted, and the experimental data was collected to establish the relationship between the application amount of quicklime and the increase in soil pH; S4, preprocessing the collected experimental data, and dividing the preprocessed data into a training set and a test set at a ratio of 8:2; S5. Use the training set to train different machine learning models, establish the relationship between quicklime application amount, key physical and chemical indicators of soil acidification, and soil pH increase, use five-fold cross-validation and external optimization validation methods to perform multiple prediction verifications, screen out the machine learning model with the best parameter combination after comparison, and adjust the model parameters; S6. Use the trained machine learning model to learn and extract key physical and chemical indicators of soil acidification, predict the results, and output the amount of quicklime required for acidic soil to reach the target pH.

2. The method for accurately predicting acid reduction in acidic cultivated soil by machine learning according to claim 1, characterized in that: Step S1 is specifically as follows: Acidic soils were collected from many provinces in my country, and the degree of soil acidification was graded according to the standards of pH < 4.5, 4.5 ≤ pH < 5.0, 5.0 ≤ pH < 5.5, and 5.5 ≤ pH < 6.

5. Soils with pH < 5.5 were selected as test soils.

3. The method for accurately predicting acid reduction in acidic cultivated soil by machine learning according to claim 1, characterized in that: In step S2, the key physical and chemical indicators of soil acidification specifically include: Soil active acid and potential acid indicators: soil pH, exchangeable aluminum, exchangeable hydrogen, total exchangeable acid; the total exchangeable acid is the sum of exchangeable aluminum and exchangeable hydrogen; Indicators of soil acid buffering capacity: soil organic matter; Soil physical indicators: soil clay.

4. The method for accurately predicting acid reduction in acidic cultivated soil by machine learning according to claim 1, characterized in that: In step S3, the soil cultivation experiment of quicklime application amount is specifically as follows: Weigh a specified mass of acidic soil, add different amounts of quicklime to the soil, mix well, and incubate at 70% saturated water holding capacity at 25°C in a sealed and light-proof place for 24 hours. Then, add carbon dioxide-free water to the soil to make the soil-water ratio of 1:2.5, stir for 5 minutes, let stand for 1 hour, and measure the pH of the supernatant.

5. The method for accurately predicting acid reduction in acidic cultivated soil by machine learning according to claim 3, characterized in that: In step S3 and step S4, the collected experimental data include: The amount of quicklime applied, key physical and chemical indicators of soil acidification, and the amount of soil pH increase.

6. The method for accurately predicting acid reduction in acidic cultivated soil by machine learning according to claim 1, characterized in that: In step S4, the preprocessing is specifically as follows: Based on the Python environment, the collected experimental data are preprocessed using the Numpy and Pandas libraries, including: Normalize the parameter units of the collected key physical and chemical index data of soil acidification to ensure data consistency; Multiple interpolation of missing values ​​was performed to maintain data integrity; After normalizing parameter units and missing values, the data is standardized as follows: Subtract the mean of each column of feature parameter values ​​from the mean of the parameter, and then divide it by the standard deviation of the feature parameter to obtain a new set of feature parameter values.

7. The method for accurately predicting acid reduction in acidic cultivated soil by machine learning according to claim 6, characterized in that: Step S5 is specifically as follows: Use training sets to train different machine learning models, including but not limited to K-nearest neighbor, random forest, XGBoost, and gradient boosted tree; Different machine learning model parameter values ​​were used in turn through five-fold cross validation and external optimization validation methods, and the determination coefficient R 2 And root mean square error RMSE to evaluate and select the best machine learning model; For the best machine learning model selected, five-fold cross validation is performed and model parameters are adjusted to continuously improve the model prediction accuracy.

8. A precise acid reduction machine learning prediction system for acidic cultivated soil, characterized by: The system adopts the machine learning prediction method for accurate acid reduction of acidic cultivated soil as described in any one of claims 1 to 7, and the system includes an acidic soil collection module, a soil physical and chemical index determination module, a quicklime application amount cultivation experiment module, an experimental data preprocessing module, a machine learning model training module and a prediction module; Acidic soil collection module, used to collect acidic soil, and classify the soil acidification degree according to the standards of pH < 4.5, 4.5 ≤ pH < 5.0, 5.0 ≤ pH < 5.5, 5.5 ≤ pH < 6.5, and select soil with pH < 5.5 as the test soil; The soil physical and chemical index determination module is used to determine the physical and chemical indexes of the collected soil and screen the key physical and chemical indexes of soil acidification; The quicklime application amount cultivation experiment module is used to conduct soil cultivation experiments on the quicklime application amount, collect experimental data and establish the relationship between the quicklime application amount and the amount of soil pH increase; The experimental data preprocessing module is used to preprocess the collected experimental data and divide the preprocessed data into a training set and a test set; The machine learning model training module uses training sets to train different machine learning models, uses five-fold cross-validation and external optimization validation methods to perform multiple prediction validations, and after comparison, selects the machine learning model with the best parameter combination, and adjusts the parameters of the model to improve the prediction accuracy; The prediction module uses the trained machine learning model to learn and extract key physical and chemical indicators of soil acidification, predict the results, and output the amount of quicklime required for acidic soil to reach the target pH.

Images