A method and system for optimizing petroleum contaminated soil leaching process based on machine learning

Abstract

The application relates to a machine learning-based oil-contaminated soil leaching process optimization method and system, which comprises the following steps: obtaining the initial pollutant concentration of the oil-contaminated soil to be repaired and a target compliance limit value; inputting the initial pollutant concentration, the target compliance limit value and a target minimum leaching efficiency into a leaching efficiency prediction model to obtain a leaching agent type and an optimal process parameter combination; the training process of the leaching efficiency prediction model comprises the following steps: obtaining oil-containing soil leaching data and performing pretreatment; performing prediction based on the pretreated data to obtain a leaching agent type and a process parameter combination prediction value; further calculating a comprehensive process benefit target function including a reagent dosage, equipment energy consumption and waste liquid generation amount, and optimizing the parameters of the prediction model until training convergence is achieved, so that a trained leaching efficiency prediction model is obtained. Compared with the prior art, the application can realize accurate prediction of the soil oil hydrocarbon removal rate and provide a reference for leaching process parameter optimization.

Description

Technical Field

[0001] This invention relates to the field of petroleum contaminated soil remediation technology, and in particular to a method and system for optimizing petroleum contaminated soil leaching process based on machine learning. Background Technology

[0002] Soil oil pollution mainly originates from leaks and emissions during oil extraction, transportation, storage, and refining. Its main pollutants are complex hydrocarbon compounds such as alkanes and aromatics, which are highly hydrophobic, difficult to degrade, and easily adsorbed onto soil particles, posing a serious threat to ecosystems and human health. For soils with high concentrations of oil contamination, chemical leaching technology is considered one of the mainstream remediation technologies due to its short remediation cycle and rapid effectiveness. However, the development and engineering application of traditional oil-bearing soil leaching processes heavily rely on empirical judgment and numerous orthogonal laboratory experiments. This "trial and error" approach is not only time-consuming and labor-intensive, consuming large amounts of reagents and resources, but it is also often limited to local optimization of a single leaching agent under specific conditions, lacking the ability for systematic horizontal comparison under multiple agents and operating conditions, and making it difficult to lock in the global optimal solution in a complex high-dimensional parameter space.

[0003] In recent years, with the rapid development of artificial intelligence technology, machine learning, due to its powerful ability to fit high-dimensional nonlinear data, has begun to be introduced into the field of soil ecological remediation. However, current research both domestically and internationally still has certain limitations. On the one hand, the research objects are relatively singular. Existing research on leaching efficiency prediction mainly focuses on heavy metal contaminated soils. For example, patent CN 120671865A discloses a method for predicting the leaching efficiency of heavy metal leaching agents in soil based on machine learning, while research on leaching prediction for soils contaminated with organic pollutants such as petroleum is extremely limited. On the other hand, the applicable remediation technologies are limited. Machine learning prediction models for petroleum hydrocarbon contaminated soils currently mainly focus on physical remediation technologies with relatively few influencing factors and simpler mechanisms. For example, Cao Yudong constructed a machine learning prediction model for the thermal desorption remediation of polycyclic aromatic hydrocarbons and petroleum hydrocarbons in soil; Zhang Shuai created a prediction model for the remediation efficiency of thermal desorption of SVOCs contaminated soil and gas-phase extraction of VOCs contaminated soil; Fang et al. also reported the application of machine learning in identifying key factors in microwave remediation of petroleum contaminated soil. The oil-bearing soil leaching system involves complex multiphase interface interactions of "soil-oil-water-surfactant", with high parameter dimensionality and strong coupling relationship. Currently, there are no research reports on predicting the leaching efficiency of oil-bearing soil based on machine learning.

[0004] Therefore, the following key issues urgently need to be addressed in the current field of oily soil leaching remediation: First, the process parameters lack systematicity and predictability. Existing leaching technology development struggles to quantify the nonlinear synergistic effects of multiple variables, such as soil physicochemical properties, leaching agent physicochemical properties (e.g., relative molecular mass, CMC, and HLB values), and experimental operating conditions, on leaching efficiency. This makes it difficult to generalize and apply laboratory results to contaminated sites of different regions and properties. Second, the predictive models are black-boxed and lack engineering guidance. The few predictive algorithms introduced into environmental engineering are mostly black-box models, lacking in-depth analysis of the importance and influence boundaries of key process parameters, and unable to provide clear, physicochemically significant guidance for on-site engineering. Third, remediation technical indicators are severely disconnected from engineering economics. Most existing technical research focuses solely on "pursuing the highest leaching efficiency," ignoring the high reagent input, high equipment energy consumption, and ecological penalties associated with high-concentration, highly active leaching agents, as well as subsequent wastewater treatment. This can easily lead to over-remediation and uncontrolled ecological regulation in actual engineering projects.

[0005] Therefore, there is an urgent need to develop an efficient, transparent, and intelligent machine learning-based optimization method for oily soil leaching processes. This method would allow for in-depth analysis of the impact mechanisms of key process parameters while ensuring compliance with environmental standards, and would enable precise optimization of leaching agents and process parameters, ultimately achieving green and low-carbon remediation of oil-contaminated soils. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a machine learning-based optimization method and system for leaching petroleum-contaminated soil.

[0007] The objective of this invention can be achieved through the following technical solutions: A machine learning-based optimization method for petroleum-contaminated soil leaching process includes: Obtain the initial contaminant concentration and target compliance limit of the oil-contaminated soil to be remediated; calculate the target minimum leaching efficiency based on the initial contaminant concentration and target compliance limit; The initial contaminant concentration, target compliance limit, and target minimum rinsing efficiency are input into a pre-built rinsing efficiency prediction model to obtain the rinsing agent type and the optimal combination of process parameters. The training process of the leaching efficiency prediction model includes: acquiring and preprocessing oily soil leaching data; constructing a dataset based on the preprocessed data and inputting it into the leaching efficiency prediction model; making predictions based on the dataset to obtain predicted values ​​for leaching agent type and process parameter combination; calculating a comprehensive process benefit objective function including reagent dosage, equipment energy consumption, and waste liquid generation based on the predicted values ​​for leaching agent type and process parameter combination, and optimizing the parameters of the leaching efficiency prediction model until training converges to obtain a trained leaching efficiency prediction model.

[0008] Furthermore, the functional expression of the comprehensive process benefit objective function includes: L = α·Dosage·V water + β·P·Time + γ·V water Where α is the normalized weighting coefficient for the dosage of the agent, and Dosage is the predicted concentration of the rinsing agent; V water γ represents the water consumption calculated based on the liquid-to-solid ratio; Time represents the predicted rinsing time; β represents the energy consumption normalization weighting coefficient; P represents the equipment operating power; γ represents the waste liquid generation normalization weighting coefficient; and L represents the comprehensive process benefit objective function.

[0009] Furthermore, the data on oily soil leaching includes soil physicochemical properties, leaching agent physicochemical properties, experimental conditions, and corresponding leaching efficiencies.

[0010] Furthermore, the soil physicochemical properties specifically include initial oil content, soil pH, soil moisture content, organic matter content, cation exchange capacity, clay percentage, and silt percentage. The physicochemical properties of the leaching agent include relative molecular mass, critical micelle concentration, and hydrophilic-lipophilic balance value. The experimental conditions include leaching agent concentration, liquid-solid ratio, stirring speed, leaching pH, leaching temperature, and leaching time.

[0011] Furthermore, the rinsing efficiency prediction model can be a category boosting model, a decision tree model, an extreme gradient boosting model, a random forest model, an artificial neural network model, or a support vector regression model.

[0012] Furthermore, the preprocessing process includes: performing unit conversion on the oily soil leaching data and removing data with missing or abnormal key features.

[0013] Furthermore, the dataset is divided into a training set and a test set; on the training set, the training hyperparameters are optimized using grid search and five-fold cross-validation.

[0014] Furthermore, the trained rinsing efficiency prediction model is evaluated using the test set; the coefficient of determination and root mean square error are used as evaluation indicators.

[0015] Furthermore, the methods also include: The SHAP algorithm is used to analyze the input features in the oil-bearing soil leaching data, and combined with Monte Carlo permutation importance verification, the core process parameters that have the greatest impact on leaching efficiency are output.

[0016] The present invention also provides a system for optimizing the leaching process of petroleum-contaminated soil based on machine learning, comprising a memory and a processor, wherein the memory stores a computer program, and the processor calls the computer program to execute the steps of any of the methods described above.

[0017] Compared with the prior art, the present invention has the following advantages: (1) This invention constructs a comprehensive process benefit objective function that simultaneously couples reagent dosage, equipment energy consumption and waste liquid generation, which can simultaneously take into account rinsing efficiency, rinsing energy consumption and water treatment burden. Traditional methods only consider rinsing efficiency, which leads to the need to use more reagents, a larger liquid-solid ratio or a longer reaction time for higher rinsing efficiency. This not only leads to a sharp increase in reagent input and equipment energy consumption, but may also significantly increase the subsequent water treatment time due to the generation of more difficult-to-treat oily wastewater, which can easily form a vicious cycle of high efficiency, high energy consumption and high burden. This invention successfully breaks through the limitations of efficiency-only optimization in traditional process optimization by embedding three mutually restrictive and even contradictory loss terms into the training process of a machine learning model as the core optimization objective. It can screen out the optimal combination of process parameters that minimizes the total operating investment while ensuring that the pollutant removal efficiency meets the remediation requirements. Thus, at the engineering application level, it achieves the organic unity of environmental compliance and process optimization, and effectively reduces the long-term operating investment of large-scale contaminated soil remediation projects.

[0018] (2) The model trained by this invention can simultaneously output the type of leaching agent and its optimal combination of process parameters, such as liquid-solid ratio, reaction time, and stirring speed. This achieves full-process coordinated prediction from agent selection to operating conditions, effectively solving the problem of mutual constraints and influences among process parameters in actual engineering, which is difficult to optimize multiple parameters simultaneously based on human experience using traditional methods. By automatically capturing the high-dimensional nonlinear relationship between initial pollutant concentration, soil properties, agent characteristics, and operating parameters through machine learning algorithms, this invention can obtain a globally coordinated and consistent optimization scheme, thereby maximizing the actual leaching remediation efficiency and avoiding the risk of a sharp drop in overall treatment effect due to improper setting of a single parameter.

[0019] (3) The model training process of this invention includes a closed-loop feedback mechanism, that is, the comprehensive process efficiency index is calculated based on the predicted leaching agent type and process parameter combination, and the model parameters are repeatedly adjusted as the optimization target until the training converges. This enables the leaching efficiency prediction model to have the ability of dynamic iteration and adaptive learning, and can continuously improve itself as training data accumulates, gradually approaching the global optimal solution. In practical applications, this model can adaptively adapt to the remediation needs of oily soils with different sources and different pollution characteristics, such as different aging degrees and different clay contents, without the need for frequent manual adjustment of logic rules or redesign of algorithms, thereby significantly improving the generalization ability and robustness of the process scheme under different working conditions. Attached Figure Description

[0020] Figure 1 This is a flowchart of a method for optimizing the leaching process of petroleum-contaminated soil based on machine learning, provided in an embodiment of the present invention. Figure 2 This is a Spearman correlation matrix diagram between any two features of a machine learning-based optimization method for leaching oil-contaminated soil provided in an embodiment of the present invention. Figure 3 This is a regression diagram of measured and predicted values ​​on the training and test sets for a machine learning model of a machine learning-based optimization method for petroleum-contaminated soil leaching process provided in an embodiment of the present invention. Figure 4 The following are examples of the CB model-based optimization method for petroleum contaminated soil leaching, as provided in this embodiment of the invention: (a) SHAP feature importance map and (b) Monte Carlo feature importance map based on the CB model. Figure 5 This is a comparison of the predicted and experimental oil removal rates (%) of the CB model for a machine learning-based optimization method for leaching oil-contaminated soil provided in this embodiment of the invention on (a) the training set and (b) the test set. Figure 6 This is a partial dependency graph of a single feature on leaching efficiency in a machine learning-based optimization method for petroleum-contaminated soil leaching process provided in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0023] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0024] Definitions: SHAP algorithm: It is a model interpretation method based on SHAP values ​​in game theory, used to measure the contribution of each feature in a machine learning model to the prediction result.

[0025] Monte Carlo permutation importance verification is a feature evaluation method that combines Monte Carlo simulation with permutation importance algorithms to robustly measure the true contribution of each feature in a machine learning model to predictive performance. Its core process involves repeatedly and independently shuffling a specific feature column on the validation set. After each shuffling, the model performance is recalculated, including the magnitude of the decrease in absolute coefficients. Finally, the importance of the feature is quantified using statistics from all repeated experiments, such as the mean and standard deviation of the performance decrease.

[0026] Example 1 like Figure 1 As shown in the figure, this embodiment provides a method for optimizing the leaching process of petroleum-contaminated soil based on machine learning. The method includes the following steps: S1: Obtain the initial contaminant concentration and target compliance limit of the oil-contaminated soil to be remediated; calculate the target minimum leaching efficiency based on the initial contaminant concentration and target compliance limit; S2: Input the initial contaminant concentration, target compliance limit and target minimum rinsing efficiency into the pre-built rinsing efficiency prediction model to obtain the rinsing agent type and the optimal combination of process parameters; S3: The training process of the leaching efficiency prediction model includes: acquiring and preprocessing oily soil leaching data; constructing a dataset based on the preprocessed data and inputting it into the leaching efficiency prediction model; making predictions based on the dataset to obtain predicted values ​​for the leaching agent type and process parameter combination; calculating the comprehensive process benefit objective function, including reagent dosage, equipment energy consumption, and waste liquid generation, based on the predicted values ​​for the leaching agent type and process parameter combination, and optimizing the parameters of the leaching efficiency prediction model until the training converges, thus obtaining the trained leaching efficiency prediction model.

[0027] S301: Data Collection and Preprocessing: Collect relevant data on oily soil leaching, including soil physicochemical properties, leaching agent physicochemical properties, experimental conditions, and corresponding leaching efficiencies, and construct an initial sample database; standardize units and handle missing values ​​in the data. In this embodiment, a total of 590 valid data samples were collected, with oil removal rate as the output feature and 14 influencing factors, including soil physicochemical properties, leaching agent properties, and leaching process conditions, as input features. Standardization methods were used to convert each feature into a standard normal distribution, and then the data was divided into training and test sets. Preferred, Soil physicochemical properties include petroleum pollution concentration, soil moisture content, soil pH, organic matter content, silt content, clay content, and cation exchange capacity; leaching agent physicochemical properties include relative molecular mass, critical micelle concentration, and hydrophilic-lipophilic balance value; experimental conditions include leaching agent concentration, liquid-solid ratio, stirring speed, leaching pH, leaching temperature, and leaching time.

[0028] Preferred, The preprocessing process specifically includes: deleting data samples with missing features and standardizing the data using Z-score or StandardScaler.

[0029] Preferred, Through literature mining and experimental data collection, an initial database was constructed covering soil physicochemical properties, pollutant occurrence characteristics, leaching agent properties, and process operation parameters. Soil physicochemical properties included: soil pH, organic matter content, cation exchange capacity, clay percentage, silt percentage, and sand percentage. Leaching agent properties included: critical micelle concentration and hydrophilic-lipophilic balance value. Process operation parameters included: leaching agent concentration, stirring speed, liquid-solid ratio, leaching temperature, leaching pH, and leaching time. The data were standardized using the StandardScaler method to eliminate dimensional differences, and the preprocessed dataset was divided into training and test sets at a ratio of 80% and 20%, respectively.

[0030] S302: Prediction Model Construction and Optimization: Construct multiple machine learning prediction models, optimize model hyperparameters on the training set using grid search and cross-validation, and establish a prediction model for oil-bearing soil leaching efficiency; evaluate the performance of each model through the test set, and select the best leaching efficiency prediction model. Specifically, Machine learning prediction models include one or more of the following: Artificial Neural Network (ANN), Category Boosting (CatBoost), Random Forest (RF), Decision Tree (DT), Extreme Gradient Boosting (XGBoost), and Support Vector Regression (SVR).

[0031] Preferred, Multi-model performance comparison and hyperparameter self-optimization: An ensemble learning candidate model was constructed, including CatBoost, Decision Tree (DT), Extreme Gradient Boosting (XGBoost), Random Forest (RF), Artificial Neural Network (ANN), and Support Vector Regression (SVR). Grid Search combined with five-fold cross-validation was introduced on the training set to automatically optimize hyperparameters such as learning rate, tree depth, and regularization coefficient for each model. The prediction accuracy of each model was evaluated using the test set, and the coefficient of determination and root mean square error were used as evaluation metrics to select the best-performing rinsing efficiency prediction model.

[0032] Preferred, Performance evaluation metrics include at least two of the following: coefficient of determination, root mean square error, mean square error, and mean absolute error.

[0033] Preferred, The training process for the rinsing efficiency prediction model includes: (1) Establish a database with the physical and chemical properties of petroleum hydrocarbon-contaminated soil, the properties of leaching agents and experimental conditions as input features, and the leaching effect, i.e. the change in petroleum hydrocarbon content before and after remediation, as the output variable; (2) Develop and optimize machine learning models for predicting the leaching efficiency of contaminated soil; (3) Based on the proposed model, key factors in the rinsing process are identified through input feature importance analysis; (4) Based on the proposed model, the selection strategy of leaching agent and the predicted value of process parameter combination for different petroleum hydrocarbon contaminated soil leaching processes are given. Based on the leaching agent type and the predicted value of process parameter combination, the comprehensive process benefit objective function including agent dosage, equipment energy consumption and waste liquid generation is calculated.

[0034] S303: Analysis of the interpretability mechanism: Based on the optimal prediction model, the algorithm is used to evaluate the contribution of input features to rinsing efficiency and the relationship with the response. Specifically, The SHAP algorithm is introduced to evaluate feature contribution, and the Monte Carlo permutation importance algorithm is combined to verify feature dependence. A partial dependency graph is plotted to analyze the nonlinear response relationship between a single parameter and rinsing efficiency.

[0035] Preferred, Multidimensional interpretability mechanism to analyze process response patterns The SHAP algorithm is used to globally interpret the optimal model and quantify the contribution of each input feature to the rinsing efficiency. The Monte Carlo permutation importance algorithm is used to verify the feature dependency. A partial dependency graph is drawn to analyze the nonlinear response boundary between core process parameters, such as rinsing agent concentration and liquid-solid ratio, and rinsing efficiency, and to identify the sensitive interval of process operation.

[0036] S304: Intelligent process optimization: Obtain the initial contaminant concentration of the soil to be remediated, set the target compliance limit, and calculate the target minimum leaching efficiency; combine the comprehensive process benefit objective function including reagent dosage, equipment energy consumption and waste liquid generation, use the best prediction model to perform batch prediction in the preset process parameter space, and output the leaching agent and process parameter combination that meets the target minimum leaching efficiency and has the lowest comprehensive process benefit.

[0037] Preferred, Intelligent optimization recommendation based on compliance constraints and economic objectives: Obtain the initial petroleum hydrocarbon concentration and post-remediation limit of the target remediation site, calculate the target leaching efficiency constraints, and construct a comprehensive process benefit objective function. L = α·Dosage·V water + β·P·Time + γ·V water Within the preset range of process parameters, the optimal prediction model is invoked for high-throughput virtual screening, automatically outputting the type of rinsing agent that meets the target rinsing efficiency and has the lowest overall process benefits, along with its optimal combination of process parameters.

[0038] Preferred, The objective function for comprehensive process benefits is: Comprehensive process benefits = Total amount of reagents added + Total energy consumption of equipment + Total amount of waste liquid generated (normalized weighted summation).

[0039] Example 2 This embodiment provides a system for optimizing the leaching process of petroleum-contaminated soil based on machine learning, including a memory and a processor. The memory stores a computer program, and the processor calls the computer program to execute the steps of the method as described in Embodiment 1.

[0040] This embodiment also provides specific implementation steps for a method for efficiently optimizing the leaching process of petroleum-contaminated soil based on machine learning, including: Step S1: Data Collection and Preprocessing Data Collection: Using keywords such as "soil remediation", "petroleum hydrocarbons (TPH)", "surfactants", and "leaching", literature was retrieved from databases such as Web of Science and CNKI to extract and construct a sample database of oil-bearing soil leaching efficiency. The database contains three categories of characteristic variables: (1) Soil physicochemical properties: initial oil content (mg / kg), soil pH, soil moisture content (%), organic matter content (%), cation exchange capacity (cmol / kg), clay percentage (%), silt percentage (%), and sand percentage (%); (2) Physicochemical properties of leaching agents: critical micelle concentration (CMC, mg / L) and hydrophilic-lipophilic balance (HLB); (3) Leaching process parameters: leaching agent concentration (g / L), stirring speed (rpm), liquid-solid ratio (mL / g), leaching temperature (°C), and leaching time (min). The target variable (label) is: oil removal rate (%).

[0041] Data cleaning and preprocessing: Unit conversion is performed on the collected data, and samples with missing or abnormal key features are removed.

[0042] Data Standardization and Splitting: To eliminate the influence of different units of measurement on model training, the StandardScaler method was used to standardize the feature data, ensuring it follows a standard normal distribution with a mean of 0 and a standard deviation of 1. Subsequently, the preprocessed dataset was randomly divided into training and test sets at a ratio of 80% to 20%. Figure 2 The figure shown is a Spearman correlation coefficient matrix between any two features.

[0043] Step S2: As Figure 3 As shown, the prediction model and optimal model construction are as follows: Six mainstream machine learning algorithms—Artificial Neural Network (ANN), CatBoost, Decision Tree (DT), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Regression (SVR)—are selected to construct the initial model. Hyperparameter optimization: On the training set, a grid search method combined with 5-fold cross-validation (CV) is used to automatically optimize the hyperparameters of each model, such as tree depth, learning rate, and number of iterations, to prevent overfitting and obtain the optimal hyperparameter combination. Figure 3 In the diagram, the corresponding models are labeled as: (a) CB, (b) RF, (c) DT, (d) XGB, (e) SVR, and (f) ANN models.

[0044] Model Evaluation and Optimization: The trained model is validated using a test set. The coefficient of determination R is used. 2 The root mean square error (RMSE) is used as an evaluation metric, and its calculation formula is as follows: in, This represents the predicted rinsing efficiency value for the i-th sample. y i This represents the measured value. denoted by , where represents the mean of the measured values, and n represents the sample size. Comparing the evaluation metrics of each model, Rm is chosen. 2 The model with the largest maximum and smallest error index RMSE is selected as the optimal model for predicting rinsing efficiency.

[0045] The evaluation metrics for different prediction models are shown in Table 1, where the R-value of the CB model on the test set is... 2 The efficiency score reached 0.898, with an RMSE of only 8.008, significantly outperforming other models. The comparison results between the oil removal rate (%) measured and predicted using the CB model on the training and test sets are as follows: Figure 4 As shown in the figure. The results confirm that the model can effectively capture the potential patterns in the experimental data and produce reliable prediction results.

[0046] Table 1: Evaluation metrics for different prediction models Step S3: Analysis of Multidimensional Interpretability Mechanisms SHAP feature importance map and Monte Carlo feature importance map based on CB model, as shown below. Figure 5 As shown. To reveal the intrinsic decision-making logic of the optimal prediction model, an interpretability algorithm is introduced to analyze the influence mechanism of process parameters from the perspective of global feature contribution: the SHAP algorithm is used to calculate the SHAP value of each input feature, and Monte Carlo ranking importance is used for cross-validation to screen out the core process parameters that have the most significant impact on rinsing efficiency. The importance ranking of each input feature is: rinsing agent concentration > rinsing time > liquid-solid ratio.

[0047] Analysis of local response patterns: such as Figure 6 As shown, a partial dependency plot is drawn to visually demonstrate the nonlinear dynamic response relationship and marginal effect between a single core parameter and leaching efficiency. Surfactant concentration (Csurfactant) is the most critical variable determining the absolute upper limit of TPH removal rate. The selection of reagents with high HLB values ​​(>15), a high liquid-to-solid ratio (LSR), and a high rotation speed (AS) provide a strong mass transfer environment, significantly generating positive marginal benefits. This technology system has a wide adaptation window to initial soil oil content (OCL) and pH, but attention should be paid to the adsorption of petroleum hydrocarbons in soils with high organic matter (OM).

[0048] Step S4: Intelligent Optimization Based on Dual Constraints of Compliance and Economy Set target constraints: Obtain the initial oil content of the target site to be repaired, set the target repair limit according to national or local environmental protection standards, and calculate the target minimum rinsing efficiency threshold.

[0049] Constructing a comprehensive process benefit objective function: Establishing a formula for calculating the comprehensive process benefit of a single rinsing treatment: L = α·Dosage·V water + β·P·Time + γ·V water In the formula: α is the normalized weighting coefficient for the dosage of the drug, and Dosage is the predicted concentration of the rinsing agent; V water γ represents the water consumption calculated based on the liquid-to-solid ratio; Time represents the predicted rinsing time; β represents the energy consumption normalization weighting coefficient; P represents the equipment operating power; γ represents the waste liquid generation normalization weighting coefficient; and L represents the comprehensive process benefit objective function.

[0050] High-throughput virtual optimization: Within a preset safety process parameter space, the system calls the optimal prediction model to perform large-scale virtual predictions. The system automatically eliminates schemes whose predicted rinsing efficiency is lower than the target threshold, and substitutes the above-mentioned comprehensive process benefit objective function into the remaining compliant schemes. Finally, it outputs the set of rinsing agent types and corresponding process parameter combinations with the lowest L as the final recommended remediation scheme.

[0051] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for optimizing the leaching process of petroleum-contaminated soil based on machine learning, characterized in that, include: Obtain the initial contaminant concentrations and target compliance limits for the oil-contaminated soil to be remediated; Calculate the target minimum rinsing efficiency based on the initial contaminant concentration and the target compliance limit; The initial contaminant concentration, target compliance limit, and target minimum rinsing efficiency are input into a pre-built rinsing efficiency prediction model to obtain the rinsing agent type and the optimal combination of process parameters. The training process of the leaching efficiency prediction model includes: acquiring and preprocessing oily soil leaching data; constructing a dataset based on the preprocessed data and inputting it into the leaching efficiency prediction model; making predictions based on the dataset to obtain predicted values ​​for leaching agent type and process parameter combination; calculating a comprehensive process benefit objective function including reagent dosage, equipment energy consumption, and waste liquid generation based on the predicted values ​​for leaching agent type and process parameter combination, and optimizing the parameters of the leaching efficiency prediction model until training converges to obtain a trained leaching efficiency prediction model.

2. The method for optimizing the leaching process of petroleum-contaminated soil based on machine learning according to claim 1, characterized in that, The functional expression of the comprehensive process benefit objective function includes: L = α·Dosage·V water + β·P·Time + γ·V water Where α is the normalized weighting coefficient for the dosage of the agent, and Dosage is the predicted concentration of the rinsing agent; V water γ represents the water consumption calculated based on the liquid-to-solid ratio; Time represents the predicted rinsing time; β represents the energy consumption normalization weighting coefficient; P represents the equipment operating power; γ represents the waste liquid generation normalization weighting coefficient; and L represents the comprehensive process benefit objective function.

3. The method for optimizing the leaching process of petroleum-contaminated soil based on machine learning according to claim 1, characterized in that, The data on oily soil leaching includes soil physicochemical properties, leaching agent physicochemical properties, experimental conditions, and corresponding leaching efficiency.

4. The method for optimizing the leaching process of petroleum-contaminated soil based on machine learning according to claim 3, characterized in that, The specific soil physicochemical properties include initial oil content, soil pH, soil moisture content, organic matter content, cation exchange capacity, clay percentage, and silt percentage. The physicochemical properties of the leaching agent include relative molecular mass, critical micelle concentration, and hydrophilic-lipophilic balance value. The experimental conditions include leaching agent concentration, liquid-solid ratio, stirring speed, leaching pH, leaching temperature, and leaching time.

5. The method for optimizing the leaching process of petroleum-contaminated soil based on machine learning according to claim 1, characterized in that, The rinsing efficiency prediction model can be a class boosting model, a decision tree model, an extreme gradient boosting model, a random forest model, an artificial neural network model, or a support vector regression model.

6. The method for optimizing the leaching process of petroleum-contaminated soil based on machine learning according to claim 1, characterized in that, The preprocessing process includes: performing unit conversion on the oil-bearing soil leaching data and removing data with missing or abnormal key features.

7. The method for optimizing the leaching process of petroleum-contaminated soil based on machine learning according to claim 1, characterized in that, The dataset is divided into a training set and a test set; on the training set, the training hyperparameters are optimized using grid search and five-fold cross-validation.

8. The method for optimizing the leaching process of petroleum-contaminated soil based on machine learning according to claim 7, characterized in that, The trained rinsing efficiency prediction model was evaluated using the test set; the coefficient of determination and root mean square error were used as evaluation indicators.

9. The method for optimizing the leaching process of petroleum-contaminated soil based on machine learning according to claim 1, characterized in that, The method further includes: The SHAP algorithm is used to analyze the input features in the oil-bearing soil leaching data, and combined with Monte Carlo permutation importance verification, the core process parameters that have the greatest impact on leaching efficiency are output.

10. A system for optimizing the leaching process of petroleum-contaminated soil based on machine learning, characterized in that, It includes a memory and a processor, the memory storing a computer program, the processor invoking the computer program to perform the steps of the method as described in any one of claims 1 to 9.

Images