Soil phosphorus adsorption capacity evaluation method based on machine learning
By establishing a soil phosphorus adsorption model using machine learning methods, the problem of time-consuming and labor-intensive laboratory assessment in existing technologies has been solved, enabling rapid and accurate assessment of soil phosphorus adsorption capacity and supporting rational fertilization and water environment protection.
Patent Information
- Application Number
- CN202510939381.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-08
- Publication Date
- 2025-10-31
AI Technical Summary
Current technologies for assessing soil phosphorus adsorption capacity rely on time-consuming and resource-intensive laboratory batch equilibrium experiments, which cannot fully reflect regional soil properties and result in limited predictions of phosphorus migration and transformation processes.
Machine learning methods were employed to establish various machine learning models by acquiring multiple sets of soil medium adsorption model parameters and physicochemical property data, optimizing hyperparameters, and predicting soil phosphorus adsorption capacity, including Langmuir and Freundlich models. Outliers were removed and data were standardized, and the optimal model was used to evaluate the regional soil phosphorus adsorption capacity.
It improves assessment efficiency and accuracy, reduces detection costs, enables rapid prediction of regional soil phosphorus adsorption capacity, supports rational fertilization and reduces phosphorus loss, and promotes water pollution control.
Smart Images
Figure CN120873591A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of environmental science and technology, and more specifically to a method for assessing soil phosphorus adsorption capacity based on machine learning. Background Technology
[0002] Human activities, such as phosphate mining, fertilizer application, and waste disposal, have significantly altered the global phosphorus cycle, leading to increased soil phosphorus content and further impacting the aquatic environment. The continuous increase in phosphorus input into water bodies may trigger eutrophication, resulting in various adverse consequences, such as algal blooms and fish deaths. Soil adsorption is a key factor influencing the migration of phosphorus from soil to the aquatic environment. Assessing soil's phosphorus adsorption capacity provides strong technical support for regional soil, surface water, and groundwater pollution risk assessment and ecological restoration, which is of great significance for maintaining ecological health and ensuring water resource security.
[0003] To date, the assessment of soil's phosphorus adsorption capacity has mainly relied on batch equilibrium experiments in the laboratory. These experimental results are used to fit soil adsorption process models such as Langmuir and Freundlich to estimate the values of model parameters.
[0004] However, these laboratory batch equilibrium experiments are not only time-consuming and resource-intensive, but also require 48 to 72 hours to determine the adsorption isotherm of a single sample, and the cost of testing can be as high as 500 to 1,000 yuan. In addition, the spatial distribution of regional soil properties is significantly heterogeneous, and the limited soil samples are difficult to fully reflect the adsorption capacity of regional soil for phosphorus, which limits the role of indoor experimental results in predicting regional phosphorus migration and transformation processes. Therefore, this invention provides a method for evaluating soil phosphorus adsorption capacity based on machine learning. Summary of the Invention
[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method for evaluating soil phosphorus adsorption capacity based on machine learning, so as to solve the problems existing in the background art.
[0006] This invention provides the following technical solution: a method for assessing soil phosphorus adsorption capacity based on machine learning, comprising the following:
[0007] S1. First, obtain multiple sets of phosphorus adsorption model parameter estimation results and soil main physicochemical property data in different types of soil media;
[0008] S2. Preprocess the acquired data;
[0009] S3. Establish multiple machine learning models to predict the main parameters of the soil phosphorus adsorption model based on the main physicochemical properties of the soil, and optimize the hyperparameters through random search cross-validation.
[0010] S4. Calculate the performance evaluation metrics of the model and select the optimal machine learning model;
[0011] S5. Apply the optimal machine learning model to evaluate the phosphorus adsorption capacity of regional soils.
[0012] Furthermore, the adsorption capacity model for phosphorus in the soil medium in S1 mainly includes the Langmuir and Freundlich models. The Langmuir adsorption isotherm formula is as follows:
[0013]
[0014] In the formula: q e Q represents the amount of adsorption at equilibrium (unit: mg / g). max : Maximum adsorption capacity of Langmuir (unit: mg / kg), K L Langmuir adsorption equilibrium constant (unit: L / mg), C e The equilibrium liquid phase concentration (unit: mg / L) is given by the Freundlich adsorption isotherm formula as follows:
[0015]
[0016] In the formula: q e Adsorption capacity at equilibrium (unit: mg / g), K F b: Freundlich adsorption capacity (unit: mg / kg), c: Freundlich adsorption constant (dimensionless), C: Freundlich adsorption capacity (unit: mg / kg) e The liquid phase concentration at equilibrium was used to obtain experimental fitting results for multiple sets of phosphorus adsorption model parameters in different soil media. For the Langmuir model, the model parameters include the maximum adsorption capacity (Q). max ) and adsorption constant (K L For the Freundlich model, its model parameters include adsorption capacity (K). F The main soil physicochemical properties that affect the phosphorus adsorption capacity of soil media include soil pH, soil particle size distribution, cation exchange capacity, and organic matter content.
[0017] Furthermore, in step 2, data points in the adsorption model parameter source data sequence that deviate from the mean by more than three standard deviations are considered outliers and removed. All source data sequences are then standardized using the formula: X scaled = (X-μ) / σ, where: μ is the mean and σ is the standard deviation. Gaussian noise with a mean of 0 and a standard deviation of σ is added to the source data of soil physicochemical properties for each group.
[0018] Furthermore, in step 3, multiple machine learning models are established to predict the main parameters of the soil phosphorus adsorption model based on the main physicochemical properties of the soil. The alternative machine learning models may include adaptive boosting, extreme gradient boosting, random forest, gradient boosting regression, support vector regression, and multilayer perceptron. A random search is used to determine the optimal hyperparameter combination for different machine learning models. Key hyperparameters of the machine learning models may include: Adaptive boosting: learning rate, number of decision trees, loss type; Extreme gradient boosting: maximum depth of decision trees, regularization coefficient, sample ratio, learning rate, number of decision trees; Random forest: maximum feature selection method, minimum number of samples per leaf node, maximum depth of decision trees, number of decision trees; Gradient boosting regression: maximum depth of decision trees, minimum number of samples required for internal node re-split, learning rate, number of decision trees; Support vector regression: penalty parameter, kernel function, kernel coefficient, tolerance error; Multilayer perceptron: number of neurons in hidden layers, activation function, regularization coefficient, initial learning rate.
[0019] Furthermore, in step 4, each machine learning model is run independently multiple times, and the mean squared error (MSE) or coefficient of determination (R²) is used to measure the results. 2 The model performance evaluation metric compares the performance of different machine learning models. The formula is: as well as In the formula: y i For the true value, For predicted values, The optimal machine learning model is determined by the mean and standard deviation of the model evaluation metrics obtained from multiple runs of the test set with different machine learning models, using the true average value.
[0020] Furthermore, the prediction function in step 5 includes acquiring regional soil types and their physicochemical properties data, applying the optimal machine learning model to calculate the phosphorus adsorption model parameter values for different types of soil, and evaluating the phosphorus adsorption capacity of regional soils.
[0021] The technical effects and advantages of this invention are as follows:
[0022] 1. This invention uses a machine learning model to quickly process large amounts of data and provide prediction results, improving evaluation efficiency. It eliminates the need for time-consuming phosphorus adsorption experiments on every new soil sample, thereby reducing sampling costs. Only key and relatively easy-to-obtain physicochemical properties need to be measured, and the adsorption parameters can be quickly predicted through a trained model, greatly improving detection speed.
[0023] 2. This invention accurately assesses soil phosphorus adsorption capacity, allowing for the determination of appropriate phosphate fertilizer application rates in subsequent agriculture, avoiding insufficient or excessive fertilization. Precise assessment of adsorption capacity effectively reduces phosphorus loss due to excessive fertilization, thereby lowering the risk of eutrophication in water bodies. Furthermore, the spatial assessment results of regional soil phosphorus adsorption capacity can assist in calibrating parameters related to phosphorus transport processes in water environment models of different regions, thus promoting effective decision-making for regional phosphorus pollution control.
[0024] 3. This invention utilizes machine learning and big data to establish a high-precision, high-efficiency, and highly generalizable prediction and evaluation method by linking easily measurable soil physicochemical properties with difficult-to-measure but crucial phosphorus adsorption model parameters. This greatly improves the accuracy of detecting soil phosphorus adsorption capacity, while also increasing the detection rate and having a wide range of applications. Attached Figure Description
[0025] Figure 1 This is a flowchart of a machine learning-based method for assessing the phosphorus adsorption capacity of regional soil in an example of the present invention.
[0026] Figure 2 The graph shows the simulation performance of six machine learning models in this invention on parameter b of the soil phosphorus Freundlich adsorption model.
[0027] Figure 3 The six machine learning models in this invention illustrate the effect of the Freundlich adsorption model parameter K on soil phosphorus. F Simulated performance graph.
[0028] Figure 4 This is a scatter plot comparing the simulated and observed values of parameter b in the soil phosphorus Freundlich adsorption model using the optimal gradient boosting regression model in this invention example.
[0029] Figure 5 The optimal gradient boosting regression model in this invention relates to the soil phosphorus Freundlich adsorption model parameter K. F Scatter plot comparing simulated and observed values.
[0030] Figure 6 The six machine learning models in this invention are used to evaluate the parameters Q of the Langmuir adsorption model. max Simulated performance graph.
[0031] Figure 7 The LogK parameter of the Langmuir adsorption model is given by six machine learning models in the examples of this invention. L Simulated performance graph.
[0032] Figure 8 The optimal gradient boosting regression model in this invention relates to the Langmuir adsorption model parameter Q. maxScatter plot comparing simulated and observed values.
[0033] Figure 9 The optimal gradient boosting regression model in this invention relates to the Langmuir adsorption model parameters LogK. L Scatter plot comparing simulated and observed values.
[0034] Figure 10 This is a graph showing the evaluation results of soil phosphorus Frendlich adsorption model parameter b on the phosphorus adsorption capacity of regional soil in an example of the present invention.
[0035] Figure 11 The parameter K in the soil phosphorus Freundlich adsorption model of this invention is... F A diagram showing the assessment results of the phosphorus adsorption capacity of regional soils.
[0036] Figure 12 The parameter Q in the Langmuir adsorption model of this invention is... max A diagram showing the assessment results of the phosphorus adsorption capacity of regional soils.
[0037] Figure 13 LogK is the parameter of the Langmuir adsorption model in the example of this invention. L A diagram showing the assessment results of the phosphorus adsorption capacity of regional soils. Detailed Implementation
[0038] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the morphology of each structure described in the following embodiments is merely illustrative. The soil phosphorus adsorption capacity assessment method based on machine learning involved in the present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0039] Example: Refer to Figures 1-13 This invention provides a method for assessing soil phosphorus adsorption capacity based on machine learning, comprising the following:
[0040] S1. First, we obtained multiple sets of phosphorus adsorption model parameter estimation results and soil physicochemical property data in different types of soil media. Specifically, we searched for domestic and foreign literature on soil phosphorus adsorption processes using keywords such as "phosphorus", "sorption", "Langmuir", and "Freundlich". From these, we selected literature that clearly listed Langmuir and Freundlich phosphorus adsorption model parameters and main soil physicochemical properties (including pH, clay content, cation exchange capacity, and organic matter content). We obtained 235 sets of phosphorus Langmuir adsorption model parameters and soil physicochemical property data and 101 sets of phosphorus Freundlich adsorption model parameters and soil physicochemical property data. Based on a large amount of literature data, we used machine learning models to mine the potential patterns in the data, avoiding the subjectivity of empirical formulas or limited datasets that traditional methods may rely on, and improving the objectivity and accuracy of the assessment.
[0041] S2. Preprocess the acquired data, including: removing data that exceeds three times the standard deviation of the mean; and processing the Langmuir model K. L The parameters were logarithmic; the data were z-standardized; four sets of Gaussian noise with a mean of 0 and a standard deviation of 0.03 were added to the standardized data for data augmentation, resulting in 1175 sets of phosphorus Langmuir adsorption model parameters and soil physicochemical property data, and 505 sets of phosphorus Freundlich adsorption model parameters and soil physicochemical property data. In the data preprocessing stage, Gaussian noise was added to augment the data, increasing the diversity of the data and helping to improve the generalization ability of the model, so that it can perform well when faced with new data or different regions and soil types.
[0042] S3. Multiple machine learning models were established to predict the main parameters of soil phosphorus adsorption models based on the main physicochemical properties of soil. Hyperparameter optimization was performed through random search cross-validation. Six machine learning models (adaptive boosting, extreme gradient boosting, random forest, gradient boosting regression, support vector regression, and multilayer perceptron) were established to predict the main parameter values of soil phosphorus Frendrich and Langmuir adsorption models based on the main physicochemical properties of soil. The data was divided into a 70% training set and a 30% test set. Five-fold cross-validation was used, and random search was employed to determine the optimal hyperparameter combination for different machine learning models. By establishing six different machine learning models and optimizing hyperparameters through random search cross-validation, optimal model performance was ensured. The diversity and optimization process helped find the most suitable model for a specific problem, improving the flexibility of evaluation. The data partitioning of 70% training set and 30% test set, five-fold cross-validation, and multiple runs with average R-values were used. 2 The evaluation ensured the reliability and stability of the model performance assessment;
[0043] S4. Calculate the performance evaluation metrics of the model, select the optimal machine learning model, run each machine learning model ten times, and calculate the R-value based on the ten runs of each model. 2 The mean is used to select the optimal machine learning model, and the model performance is evaluated using metrics such as R-squared. 2 By comparing the predicted values, the gradient boosting regression model with the highest prediction accuracy was selected. Its high coefficient of determination indicates that the model's predicted values are in good agreement with the actual observed values, thus improving the reliability of the evaluation results.
[0044] S5. Apply the optimal machine learning model to evaluate the phosphorus adsorption capacity of regional soils and obtain the distribution map of soil types in a certain region and the physicochemical properties (pH, clay content, cation exchange capacity, organic matter content) data of different soils.
[0045] The adsorption capacity model for phosphorus in the soil medium in S1 mainly includes the Langmuir and Freundlich models. The Langmuir adsorption isotherm formula is as follows:
[0046]
[0047] In the formula: q e Q represents the amount of adsorption at equilibrium (unit: mg / g). max : Maximum adsorption capacity of Langmuir (unit: mg / kg), K L Langmuir adsorption equilibrium constant (unit: L / mg), C e The equilibrium liquid phase concentration (unit: mg / L) is given by the Freundlich adsorption isotherm formula as follows:
[0048]
[0049] In the formula: q e K represents the amount of adsorption at equilibrium (unit: mg / g). F b represents the Freundlich adsorption capacity (unit: mg / kg), b represents the Freundlich adsorption constant (dimensionless), and C represents the Freundlich adsorption capacity (unit: mg / kg). e The liquid phase concentration at equilibrium is represented by the experimental fitting results of multiple sets of phosphorus adsorption model parameters in different soil media. For the Langmuir model, the model parameters include the maximum adsorption capacity (Q). max ) and adsorption constant (K L For the Freundlich model, its model parameters include adsorption capacity (K). F The main soil physicochemical properties that affect the phosphorus adsorption capacity of soil media include soil pH, soil particle size distribution, cation exchange capacity, and organic matter content.
[0050] In step 2, data points in the adsorption model parameter source data sequence that deviate from the mean by more than three standard deviations are considered outliers and removed. All source data sequences are then standardized using the formula: X scaled = (X-μ) / σ, where: μ is the mean and σ is the standard deviation. Gaussian noise with a mean of 0 and a standard deviation of σ is added to the source data of soil physicochemical properties for each group.
[0051] In step 3, multiple machine learning models are established to predict the main parameters of the soil phosphorus adsorption model based on the main physicochemical properties of the soil. The alternative machine learning models may include adaptive boosting, extreme gradient boosting, random forest, gradient boosting regression, support vector regression, and multilayer perceptron. A random search is used to determine the optimal hyperparameter combination for different machine learning models. Key hyperparameters of the machine learning models may include: Adaptive boosting: learning rate, number of decision trees, loss type, etc.; Extreme gradient boosting: maximum depth of decision trees, regularization coefficient, sample ratio, learning rate, number of decision trees; Random forest: maximum feature selection method, minimum number of samples per leaf node, maximum depth of decision trees, number of decision trees, etc.; Gradient boosting regression: maximum depth of decision trees, minimum number of samples required for internal node re-partitioning, learning rate, number of decision trees, etc.; Support vector regression: penalty parameter, kernel function, kernel coefficient, tolerance error; Multilayer perceptron: number of hidden layer neurons, activation function, regularization coefficient, initial learning.
[0052] In step 4, each machine learning model is run independently multiple times, and the mean squared error (MSE) or coefficient of determination (R²) is used to measure the results. 2 The model performance evaluation metric compares the performance of different machine learning models. The formula is: as well as In the formula: y i For the true value, For predicted values, The optimal machine learning model is determined by the mean and standard deviation of the model evaluation metrics obtained from multiple runs of the test set with different machine learning models, using the true average value.
[0053] The prediction function in step 5 includes acquiring regional soil types and their physicochemical properties data, applying the optimal machine learning model to calculate the phosphorus adsorption model parameter values for different soil types, and evaluating the phosphorus adsorption capacity of regional soils. The process adopts data acquisition-preprocessing-model construction and optimization-evaluation-application, which is clear, standardized, easy to automate and extend to other adsorption models or soil property prediction. This method is not only applicable to specific regions or soil types, but can also be applied to different regions and soil types. By acquiring regional soil types and their physicochemical properties data, the phosphorus adsorption capacity of the soil in that region can be evaluated, which helps to formulate more reasonable fertilization strategies, reduce phosphorus loss and environmental pollution, and has application prospects and practicality. At the same time, the spatial evaluation results of regional soil phosphorus adsorption capacity can also help to calibrate the relevant parameters of phosphorus transport processes in different regional water environment models, thereby promoting effective decision-making for regional phosphorus pollution control.
[0054] like Figures 2-5 Six machine learning models were used to analyze the parameters b and K of the Freundlich adsorption model. F The simulation performance of each model for the Freundlich adsorption model parameter b was ranked as follows: Gradient Boosting Regression (0.923), Random Forest (0.910), Extreme Gradient Boosting (0.898), Multilayer Perceptron (0.885), Support Vector Regression (0.862), and Adaptive Boosting (0.784). The simulation performance of each model for the Freundlich adsorption model parameter K was also ranked as follows. F The simulation performance ranking is as follows: Gradient Boosting Regression (0.942), Extreme Gradient Boosting (0.923), Random Forest (0.916), Multilayer Perceptron (0.880), Support Vector Regression (0.859), and Adaptive Boosting (0.758). Simulation results also show the optimal gradient boosting regression model for the Freundlich adsorption model parameters b and K on the test set. F The simulated and observed values were compared, and the coefficients of determination for the optimal model were found to be 0.967 and 0.964, respectively.
[0055] like Figures 6-9 Meanwhile, the parameters Q of the Langmuir adsorption model were analyzed using six machine learning models. max and LogK L The simulation performance of each model, and the parameters Q of the Langmuir adsorption model. max The simulation performance ranking is as follows: Gradient Boosting Regression (0.750), Extreme Gradient Boosting (0.715), Random Forest (0.675), Multilayer Perceptron (0.531), Support Vector Regression (0.519), and Adaptive Boosting (0.309); the simulation parameters logK for each model are as follows: LThe simulation performance ranking is as follows: Gradient Boosting Regression (0.834), Extreme Gradient Boosting (0.789), Random Forest (0.730), Support Vector Regression (0.575), Multilayer Perceptron (0.572), and Adaptive Boosting (0.353). The optimal gradient boosting regression model for the Langmuir adsorption model parameter Q is also given. max and LogK L The scatter plot comparing the simulated and observed values shows that the coefficients of determination for the optimal model are 0.861 and 0.901, respectively.
[0056] The table below provides the physicochemical properties of the main soil types in this region.
[0057]
[0058]
[0059] The optimal gradient lifting regression model was used to calculate the parameter values of the Freundlich and Langmuir adsorption models for different soils, and the spatial distribution of the parameters of the two adsorption models was obtained. Figure 10 The graph represents the Freundlich adsorption model parameter *b*, visually illustrating its spatial variation within the region using different colors. Red and orange areas are concentrated in the center of the graph, indicating higher *b* values in these areas, implying stronger adsorption capacity. Light blue and blue-green areas are scattered at the edges of the graph, representing lower *b* values and relatively weaker adsorption capacity in these areas. Figure 11 The parameter K represents the Freundlich adsorption model parameter. F The red areas in the image indicate that these regions have a high adsorption capacity for pollutants, suggesting greater potential for pollution control and remediation. This image visually shows the different K values in each region. F The distribution of K values can be analyzed to determine the differences in adsorption capacity for pollutants in different regions. This information can then be used in pollution control and remediation work. F For the distribution of values, K is preferred. F Areas with high K values can undergo natural purification, or K can be targeted. F In areas with lower adsorption values, appropriate measures should be taken to enhance adsorption. Figure 12 The value represents the Langmuir adsorption model parameter Q. max Q is displayed intuitively through color changes. max The distribution of Q is shown, with colors gradation from yellow to red, and red areas representing higher Q values. max The numerical values indicate that the soil in these areas has a higher adsorption capacity; while the yellow areas represent lower Q values. max The numerical value indicates a relatively low adsorption capacity. Analysis of Q... maxThe distribution of Q can identify areas with high adsorption capacity, facilitating the development of targeted environmental remediation strategies and optimization of pollutant control measures. Therefore, Q can be prioritized. max For areas with high values, natural attenuation or monitored natural attenuation (MNA) should be performed to reduce repair costs. Figure 13 The value represents the parameter K in the Langmuir adsorption model. L The color differences in different regions of the image clearly demonstrate the spatial variation of adsorption affinity, especially at high K values. L The yellow area of the value is related to low K. L The contrast of the blue areas indicates that the soils in different regions have different adsorption affinities, with high K values showing greater affinity. L The yellow area with a high K value indicates that the region has a high affinity for adsorption and is more likely to adsorb phosphorus. Conversely, a low K value indicates a lower affinity for phosphorus. L The blue area of the value has weak adsorption affinity. In agriculture, this can be determined based on the soil's K... L This is to optimize the use and management of fertilizers.
[0060] In summary, the spatial layout approach based on the model results can quickly and intuitively reflect the local soil adsorption situation, greatly reducing the difficulty of detection, reflecting the differences in adsorption capacity in different regions, and providing valuable information for environmental applications and engineering practices.
[0061] It will be apparent to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.
[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for assessing soil phosphorus adsorption capacity based on machine learning, characterized in that: This includes the following: S1. First, obtain multiple sets of phosphorus adsorption model parameter estimation results and soil main physicochemical property data in different types of soil media; S2. Preprocess the acquired data; S3. Establish multiple machine learning models to predict the main parameters of the soil phosphorus adsorption model based on the main physicochemical properties of the soil, and optimize the hyperparameters through random search cross-validation. S4. Calculate the performance evaluation metrics of the model and select the optimal machine learning model; S5. Apply the optimal machine learning model to evaluate the phosphorus adsorption capacity of regional soils.
2. The method for evaluating soil phosphorus adsorption capacity based on machine learning according to claim 1, characterized in that: The soil medium's phosphorus adsorption capacity model in S1 mainly includes the Langmuir and Freundlich models. The Langmuir adsorption isotherm formula is as follows: In the formula: q e Q represents the amount of adsorption at equilibrium (unit: mg / g). max : Maximum adsorption capacity of Langmuir (unit: mg / kg), K L Langmuir adsorption equilibrium constant (unit: L / mg), C e The equilibrium liquid phase concentration (unit: mg / L) is given by the Freundlich adsorption isotherm formula as follows: In the formula: q e Adsorption capacity at equilibrium (unit: mg / g), K F b: Freundlich adsorption capacity (unit: mg / kg), c: Freundlich adsorption constant (dimensionless), C: Freundlich adsorption capacity (unit: mg / kg) e The liquid phase concentration at equilibrium was used to obtain experimental fitting results for multiple sets of phosphorus adsorption model parameters in different soil media. For the Langmuir model, the model parameters include the maximum adsorption capacity (Q). max ) and adsorption constant (K L For the Freundlich model, its model parameters include adsorption capacity (K). F The main soil physicochemical properties that affect the phosphorus adsorption capacity of soil media include soil pH, soil particle size distribution, cation exchange capacity, and organic matter content.
3. The method for evaluating soil phosphorus adsorption capacity based on machine learning according to claim 1, characterized in that: In step 2, data points in the adsorption model parameter source data sequence that deviate from the mean by more than three standard deviations are considered outliers and removed. All source data sequences are then standardized using the formula: X scaled = (X-μ) / σ, where: μ is the mean and σ is the standard deviation. Gaussian noise with a mean of 0 and a standard deviation of σ is added to the source data of soil physicochemical properties for each group.
4. The method for evaluating soil phosphorus adsorption capacity based on machine learning according to claim 1, characterized in that: In step 3, multiple machine learning models are established to predict the main parameters of the soil phosphorus adsorption model based on the main physicochemical properties of the soil. The alternative machine learning models may include adaptive boosting, extreme gradient boosting, random forest, gradient boosting regression, support vector regression, and multilayer perceptron. A random search is used to determine the optimal hyperparameter combination for different machine learning models. Key hyperparameters of the machine learning models may include: Adaptive boosting: learning rate, number of decision trees, loss type; Extreme gradient boosting: maximum depth of decision trees, regularization coefficient, sample ratio, learning rate, number of decision trees; Random forest: maximum feature selection method, minimum number of samples per leaf node, maximum depth of decision trees, number of decision trees; Gradient boosting regression: maximum depth of decision trees, minimum number of samples required for internal node re-split, learning rate, number of decision trees; Support vector regression: penalty parameter, kernel function, kernel coefficient, tolerance error; Multilayer perceptron: number of neurons in hidden layers, activation function, regularization coefficient, initial learning rate.
5. The method for evaluating soil phosphorus adsorption capacity based on machine learning according to claim 1, characterized in that: In step 4, each machine learning model is run independently multiple times, based on the coefficient of determination (R²). 2 The model performance evaluation metric compares the performance of different machine learning models. The formula is: In the formula: y i For the true value, For predicted values, The optimal machine learning model is determined by the mean and standard deviation of the model evaluation metrics obtained from multiple runs of the test set with different machine learning models, using the true average value.
6. The method for evaluating soil phosphorus adsorption capacity based on machine learning according to claim 1, characterized in that: The prediction function in step 5 includes acquiring regional soil types and their physicochemical properties data, applying the optimal machine learning model to calculate the phosphorus adsorption model parameter values for different types of soil, and evaluating the phosphorus adsorption capacity of regional soils.