Ecological restoration method for heavy metal pollution in river and lake water sediments based on BN model

Abstract

The invention belongs to the technical field of water environment treatment and ecological restoration, and particularly discloses a BN model-based ecological restoration method for heavy metal pollution in river and lake water sediments. According to the method, the synergistic effects of key environmental factors such as CEC, OM, SFL, DO, TN and TP are creatively integrated, a BN model capable of accurately simulating the adsorption-desorption process of heavy metal on the water-sediment interface is constructed, the model is used for reasoning and analyzing the heavy metal migration and transformation rule, the repair strategy is optimized, and the effect is verified. According to the method, the remediation efficiency is remarkably improved, the risk of secondary pollution is reduced, and scientific and technical support is provided for heavy metal treatment in complex river and lake environments.

Description

Technical Field

[0001] This invention belongs to the field of water environment management and ecological restoration technology, specifically involving an ecological restoration method for heavy metal pollution in river and lake sediments based on the BN model. Background Technology

[0002] Heavy metals, as major environmental pollutants, accumulate in lake sediments through adsorption, sedimentation, or flocculation after entering the lake, thus purifying the overlying water to some extent. However, when environmental conditions change, heavy metals in the sediments can be released again through suspension, desorption, and various transformations, leading to "secondary pollution" of the water. Because heavy metals are difficult to biodegrade and easily bioaccumulate and amplify through the food chain, they may ultimately endanger human health and survival.

[0003] Currently, remediation methods for heavy metals in river and lake sediments mainly include physical remediation, chemical remediation, and bioremediation. While these traditional methods are effective in certain situations, they have significant limitations: physical remediation (such as dredging) is costly and prone to causing secondary disturbances; chemical remediation (such as adding chelating agents) can easily alter the chemical balance of the water body; and bioremediation has a long cycle and is significantly constrained by environmental conditions. Furthermore, traditional methods often focus on the impact of single environmental factors, neglecting the complex regulation of heavy metal migration and transformation through multi-factor interactions, making them ill-suited to address the dynamically changing river and lake environments.

[0004] In recent years, Bayesian Network (BN) models have been applied to the field of environmental pollution, and their excellent causal reasoning ability and advantages in handling complex variable relationships have made them suitable for the monitoring and assessment of heavy metal pollution. However, existing models have limitations in variable selection and lack targeted remediation strategy outputs. For example, Wijesiri et al. (2019b) proposed a method to investigate the interaction between heavy metals and water and sediments using a BN model, but the variables in this model only included organic carbon, pH value, and the mineral composition of sediments. Lorena et al. (2021) improved the BN model by considering the above variables and adding variables such as conductivity, CEC, SSA, and binding competition between heavy metals. However, factors such as potassium permanganate index, dissolved oxygen (DO), chemical oxygen demand (COD), total nitrogen (TN), total phosphorus (TP), and NH3, which can significantly alter the complexing capacity and ionic properties of heavy metals in water bodies, were not considered. Yin et al. (2024) used the BN model to assess heavy metal pollution in marine environments. This method primarily relies on remote sensing data and some physicochemical factors, failing to cover other important environmental factors in inland river and lake environments, such as dissolved oxygen, COD, and nitrogen and phosphorus concentrations, which significantly influence the adsorption-desorption behavior of heavy metals. Furthermore, this model is mainly applicable to marine environments and cannot effectively address the dynamic changes in inland river and lake water bodies, especially in estuaries where the hydrological and chemical environments are more complex.

[0005] Current research on the behavior of heavy metals in river and lake sediments is mostly descriptive, with insufficient in-depth study of the re-release of heavy metals from sediments. Due to the complexity of environmental changes, quantitative research is challenging, which is precisely one of the main reasons for "secondary pollution" of aquatic ecosystems. Therefore, there is an urgent need to construct a Bayesian network method that can integrate multiple environmental factors, accurately simulate the interfacial behavior of heavy metals, and directly guide remediation practices. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the purpose of this invention is to provide an ecological restoration method based on the BN model. By integrating the synergistic effects of key environmental factors such as CEC, OM, SFL, DO, TN, and TP, a BN model is constructed to accurately simulate the adsorption-desorption process of heavy metals at the water-sediment interface. Based on this model and utilizing its reasoning mechanism, restoration strategies are optimized to improve restoration efficiency.

[0007] The BN model described in this invention is a directed acyclic graph consisting of variable nodes and directed edges. Node variables include heavy metals in water, heavy metals in sediments, and key environmental factors (such as pH, TN, TP, DOM, OM, CEC, DO, SS, etc.). Directed edges represent causal relationships, with "+" indicating a positive effect and "-" indicating a negative effect, thus clearly expressing the promoting or inhibiting effects of each environmental factor on heavy metal migration and transformation. This model structure can comprehensively reflect the interaction between heavy metals and their controlling factors at the water-sediment interface.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: An ecological remediation method for heavy metal pollution in river and lake sediments based on the BN model, the method comprising the following steps: (1) Collect sediment and overlying water samples in typical estuary and lake areas of the target watershed, determine the concentration of heavy metals in the samples and the environmental factors affecting the migration and transformation of heavy metals, and preprocess the data. (The preprocessed data serves as input for building the BN model, and is used for subsequent node variable setting, causal structure learning, and conditional probability distribution estimation.) (2) With the goal of reducing the concentration of heavy metals in the water body and sediment to be tested, the node variables of the BN model are set based on the preprocessed data obtained in step (1). (3) Combining literature, expert knowledge and sensitivity analysis of existing data, determine the causal relationship between node variables, construct the initial BN network structure, and use "+" and "-" to represent positive and negative effects to label each causal edge, forming a structure diagram containing both causal direction and effect direction information; (4) The optimal structure of the BN model is selected through parameter learning and structure learning. After iterative optimization by the model verification program, a stable, efficient BN model with environmental mechanism support is finally constructed. (5) Based on the BN model constructed in step (4), its reasoning mechanism is used to deduce the concentration change trend of heavy metals in the water body and sediment to be tested after inputting specific environmental factor values ​​or states, identify the key environmental factors with the greatest impact and clear direction, and formulate pollution remediation measures accordingly. (6) After the remediation measures are implemented, monitoring data are continuously collected and the status of key nodes of the BN model is dynamically updated. By comparing and analyzing the changes in bioavailable heavy metal content and key environmental factors in water and sediment before and after remediation, the remediation effect is verified and the strategy is optimized and iterated.

[0009] Furthermore, the generalized graph of the BN model with environmental mechanism support capability described in step (4) is a directed acyclic graph composed of multiple variable nodes. The nodes are connected by directed edges, and "+" and "-" symbols are marked on the edges to indicate the positive and negative effects between variables, so as to describe the dual information of causal direction and effect direction at the same time. The model can describe the interaction between metals and control factors between water and sediment interfaces in all inland river and lake water bodies.

[0010] Furthermore, the heavy metals mentioned in step (1) include Cr, Cd, Cu, Zn, As, Hg, Pb and other common heavy metals; wherein, the heavy metal concentration is determined by speciation analysis, including the detection of exchangeable, organically bound and other bioavailable forms.

[0011] Furthermore, the key environmental factors mentioned in step (1) include pH, total nitrogen (TN), total phosphorus (TP), dissolved organic matter (DOM), temperature (TEMP), and chloride ions (Cl). - The parameters to be considered include dissolved oxygen (DO), cation exchange capacity (CEC), organic matter (OM), suspended solids (SS), and soluble fluoride (SFL). Furthermore, sediment-related characteristic parameters such as soil particle size and mineral composition, specific surface area (SSA), characteristic pollutant indicators such as anionic surfactant (LAS), and interfacial process parameters such as microbial activity intensity (MAI) and sulfide oxidation rate (SOR) can be combined to comprehensively characterize water-sediment interface conditions and their impact on heavy metal migration and transformation.

[0012] Furthermore, the data preprocessing in step (1) includes outlier handling, missing value imputation, and normalization; wherein: outlier handling adopts Z-score method combined with adjacent period and spatial nearest neighbor method for correction or removal, missing value imputation adopts multiple interpolation method (MICE), and normalization adopts Z-score standardization or min-max standardization method.

[0013] Furthermore, the node variables in step (2) include heavy metals in water bodies, heavy metals in sediments, and key environmental factors that affect their migration and transformation, such as pH, TN, TP, DOM, OM, CEC, TEMP, DO, CL, SS, LAS, SSA, MAI, and SOR.

[0014] Furthermore, the labeling of positive and negative effects in step (3) is based on the actual environmental interaction mechanism between variables, after determining the node connection relationship. The "+" represents a positive effect, indicating that the increase of the variable will enhance the level of the target variable, and the "-" represents a negative effect, indicating that the increase of the variable will inhibit the level of the target variable. For example, organic matter (OM) enhances the adsorption capacity of sediments for heavy metals through complexation, and is labeled as "+"; while suspended matter (SS) may reduce the adsorption of sediments for heavy metals by competing for adsorption sites, and is labeled as "-".

[0015] Furthermore, the parameter learning in step (4) specifically involves: for discrete nodes, counting the frequency of each state of the child nodes under different combinations of parent nodes, and calculating the conditional probability table in combination with Bayesian prior information; for continuous nodes, assuming that they follow a normal distribution, using the expectation-maximization algorithm to estimate their conditional distribution parameters.

[0016] Furthermore, the structure learning in step (4) specifically involves using the Bayesian Information Criterion (BIC) as the scoring function and selecting the network model with the optimal structure through heuristic search operations such as adding, deleting, and reversing edges.

[0017] Furthermore, the model validation in step (4) includes two parts: cross-validation and expert validation. Cross-validation involves dividing the data into a training set (X%) and a test set (1-X%), and using indicators such as root mean square error (RMSE) and coefficient of determination (R²) to evaluate the model performance. Expert validation is conducted by experts in the fields of environmental science and environmental chemistry to evaluate the model based on the rationality of the model structure and reasoning logic. If the model prediction error is large or the structure is unreasonable, the data processing, causal structure, or parameter learning process is adjusted back until a stable, efficient Bayesian network model with environmental mechanism support is formed.

[0018] Furthermore, the method also includes the following steps: Based on the completed BN model, its reasoning mechanism can be used to infer the concentration change trend of heavy metals in the water body and sediment under test after inputting specific environmental factor values ​​or states, identify the key factors with the greatest impact and clear direction, and provide a basis for the selection of pollution remediation strategies. After the remediation measures are implemented, monitoring data are continuously collected and the status of key nodes in the BN model is dynamically updated. Based on the comparative analysis of changes in bioavailable heavy metal content and key factors in water and sediment before and after remediation, the remediation effectiveness is verified and the strategy is optimized and iterated.

[0019] Furthermore, the remediation strategies include one or more combinations of physical remediation (such as dredging and covering), chemical remediation (such as adding precipitants and oxidants), and bioremediation (such as phytoremediation and microbial enhancement).

[0020] This invention provides an ecological remediation method for heavy metal pollution in river and lake sediments based on a Bayesian network (BN) model. This method creatively integrates the synergistic effects of key environmental factors into a Bayesian network to construct a BN model that accurately simulates the adsorption-desorption process of heavy metals at the water-sediment interface. Using this model, not only can the influence of the physicochemical properties of river and lake water and sediments on the heavy metal exchange behavior at the interface be analyzed rapidly and comprehensively, but the competitive relationships among environmental factors can also be elucidated, thereby enabling quantitative research on the water-sediment interface. Compared with existing technologies, the advantages of this invention are as follows: First, in response to the problem that the migration and transformation of heavy metals in river and lake sediments are dynamically influenced by multiple factors such as water temperature, pH value, and dissolved oxygen, and that traditional methods are difficult to handle such complex relationships, this invention efficiently integrates multi-dimensional data associations through the BN model and accurately identifies pollutant migration and transformation paths with the help of causal reasoning, thus successfully overcoming the technical obstacles brought about by data complexity and scarcity. Secondly, addressing the limitations of existing remediation methods that can only consider a few environmental factors and cannot systematically analyze the competition and synergistic effects among multiple factors, this invention, based on the structural characteristics of the BN model, achieves a comprehensive analysis of the interactions among multiple factors such as the physicochemical properties of water bodies, the types and concentrations of heavy metals, providing comprehensive support for the optimization of remediation strategies.

[0021] Furthermore, addressing the issues of traditional remediation methods relying on experience and struggling to dynamically adapt to diverse environmental conditions, resulting in unstable remediation effects and high costs, this invention utilizes a Bayesian inference mechanism to accurately assess whether ecological remediation work is necessary for heavy metals in river and lake water and sediments through conventional ecological and environmental monitoring indicators. It also automatically optimizes the remediation plan based on different remediation scenarios, significantly improving remediation efficiency and reducing implementation costs. Attached Figure Description

[0022] Figure 1 A generalized diagram of the BN model to describe the interaction between metals and controlling factors at the water-sediment interface. Detailed Implementation

[0023] Example 1: Remediation of Heavy Metal Pollution in Sediments of a Lake (1) Data acquisition and processing.

[0024] Sampling: A lake contaminated with heavy metals was selected as the research subject. Ten representative sampling points were set up in the estuary and lake areas. At each sampling point, surface (0-20cm), middle (20-50cm), and bottom (50-80cm) sediment samples, as well as overlying water samples at the corresponding depths, were collected. Sediment samples from multiple periods of river and lake water were collected, totaling 30 sediment samples and 30 overlying water samples. (Sampling points were set up in both the estuary and the central lake area to consider the different impacts of unidirectional and bidirectional flow on the migration paths of heavy metals under hydrodynamic disturbances. If the sampling period was monthly, time-series data for at least 6 periods were continuously obtained to ensure data coverage of different environmental conditions.)

[0025] Detection: The heavy metals measured in the sediment and overlying water in this embodiment were Cr, Cd, Cu, Zn, As, Hg, and Pb. The heavy metal concentration was determined using speciation analysis, covering five forms: exchangeable, carbonate-bound, iron-manganese oxide-bound, organic-bound, and residual, to estimate the bioavailable content.

[0026] The environmental factors measured in this embodiment include pH, total nitrogen (TN), total phosphorus (TP), dissolved organic matter (DOM), temperature (TEMP), and chloride ions (Cl). - Typical aquatic environmental factors such as dissolved oxygen (DO), cation exchange capacity (CEC), organic matter (OM), suspended solids (SS), water temperature (WT), and soluble fluoride (SFL) are used to comprehensively characterize water-sediment interface conditions and their impact on heavy metal migration and transformation. These parameters are combined with sediment-related characteristic parameters such as soil particle size and mineral composition, specific surface area (SSA), characteristic pollutant indicators such as anionic surface activity (LAS), and interfacial process parameters such as microbial activity intensity (MAI) and sulfide oxidation rate (SOR).

[0027] The determination of the environmental factors can be carried out according to national standard methods or commonly used technical routes in literature. Commonly used determination methods include, but are not limited to: pH measured directly with a pH meter; TN measured using the Kjeldahl method or alkaline potassium persulfate oxidation ultraviolet spectrophotometry; TP measured using ammonium molybdate spectrophotometry; DOM analyzed by solid-phase extraction combined with ultraviolet-visible spectroscopy; WT or TEMP measured using a temperature sensor or multi-parameter water quality probe; CL measured using ion chromatography or potentiometric titration; DO measured using a membrane electrode method or fluorescence dissolved oxygen meter; and CEC measured using ammonium acetate. The exchange method or trichloroacetic acid extraction method was used to determine OM, and the potassium dichromate-sulfuric acid oxidation method (Walkley-Black method) was used to determine OM. Soil properties, including particle size and mineral composition, were analyzed by laser particle size analyzer and X-ray diffractometer (XRD), respectively. SS was determined by gravimetric method or filter membrane weighing method. LAS was determined by methylene blue spectrophotometry. SSA was determined by BET method using a specific surface area analyzer. MAI was measured by enzyme activity assay or gene expression detection, etc., and SOR was calculated by constant temperature and humidity chamber method.

[0028] Preprocessing: The collected data is inspected, including outlier handling, missing value imputation, and standardization, to improve data quality, reduce noise interference, and ensure the stability and accuracy of BN modeling.

[0029] Outlier handling employs the Z-score method, calculating the standard deviation range for each variable, identifying data points exceeding a set threshold, and manually reviewing and correcting or removing outliers based on the numerical distribution of adjacent periods and spatially nearby sampling points. Missing value imputation uses the Multiple Imputation and Interpolation (MICE) method, constructing a regression model among multiple variables to iteratively predict and impute missing units multiple times, generating a complete dataset conforming to the joint distribution. Data standardization employs either min-max standardization or Z-score standardization: for variables with small distribution skewness, Z-score standardization is preferred to maintain a mean of 0 and a standard deviation of 1; for variables with large dimensional differences and no normality requirement, min-max standardization is preferred to compress them to the [0,1] interval. The preprocessed data serves as the input dataset for node variables in the BN model, used in subsequent structure learning and parameter estimation processes.

[0030] In this embodiment, two sediment samples were found to have abnormal cadmium content data and were therefore removed. Missing values ​​were filled using a random forest algorithm.

[0031] The preprocessed data serves as input for building the BN model, and is used for subsequent node variable setting, causal structure learning, and conditional probability distribution estimation.

[0032] (2) Based on the preprocessed data and repair objectives in step (1), determine the node variables of the BN model.

[0033] The node variables in this embodiment include heavy metals in water, heavy metals in sediments, and environmental factors that affect their migration and transformation, such as pH, TN, TP, DOM, OM, CEC, TEMP, DO, CL, SS, LAS, SSA, MAI, and SOR.

[0034] (3) Combine literature, expert knowledge and sensitivity analysis of existing data to determine the causal relationship between node variables, construct the initial BN network structure, and use "+" and "-" to represent positive and negative effects to label each causal edge, forming a structure diagram containing both causal direction and effect direction information.

[0035] Specifically, in this embodiment, the determination of the causal relationships and their positive and negative effects among the variables at each node adopts a structure prior construction method based on multi-source information fusion, which specifically includes the following three steps: First, through systematic literature retrieval and analysis, a causal relationship path comparison library for the migration and adsorption of heavy metals in sediments was established. Typical combinations of causal variables and their influence directions in existing studies were summarized (e.g., OM→Pb is a promoting relationship, DO→Cd is an inhibiting relationship, etc.). A literature knowledge base was constructed as the prior input for the model. In this embodiment, the selected literature is as follows: Atkinson et al. (2007) found that the mass concentration of dissolved oxygen (DO) is a necessary condition for the adsorption and release of heavy metals (HMs) at the water-sediment interface; Wang et al. (2004) found that the adsorption of phosphate increases the negative charge on the adsorbent surface and reduces the increase in surface potential, thereby increasing the adsorption of HMs; Juang et al. (2004) concluded by studying the adsorption of Cu, Zn, and P on FeOOH that P and Cu can mutually promote adsorption due to the formation of a three-layer complex in the form of FeOOH-Cu-PO4 between the adsorbents; Dijkstra et al. (2004) found that pH determines the protonation process of functional groups on the surface of the environmental medium, which is the main reason for the change in surface charge of the medium and will indirectly affect the adsorption / desorption and precipitation / dissolution equilibrium of heavy metals at the environmental medium interface; Mak and Du et al. (2009; 2011) found that organic matter has a high complexing ability for heavy metal ions based on its functional groups and negative charge. Furthermore, in the presence of heavy metal ions, organometallic complexes are easily formed, leading to changes in the biogeochemical behavior of heavy metal ions. For example, DOM can form solid complexes with HM. In addition, Williams and Rice (1932) showed that soil organic matter content increases cation exchange capacity, while Wright and Foss (1968) found a correlation between organic matter content and CEC.

[0036] Secondly, based on the modeling experience of experts in the field of water environment and ecological restoration, a set of action mechanism rules was extracted to clarify the possible direct / indirect action mechanisms between variables (such as "higher CEC usually enhances the adsorption capacity of metal cations" and "high concentration of SS may compete for adsorption sites and produce colloidal shielding effect"). These rules were then used as a template for preliminary screening of causal edges. Finally, based on the preprocessed multi-period monitoring data, the relationships between variables were quantitatively verified. Spearman rank correlation analysis, mutual information analysis, or univariate sensitivity analysis were used to assess the significance and directional consistency of causal paths. Paths supported by literature, rule bases, and data analysis were retained; conflicting paths were annotated and revised according to rule priority and statistical significance principles.

[0037] Based on the established node connectivity, directed edges are labeled with positive or negative effects according to the actual environmental interaction mechanisms between variables: a positive effect indicates that increasing the variable will enhance the level of the target variable, while a negative effect indicates that increasing the variable will inhibit the level of the target variable. For example, organic matter (OM) promotes the adsorption of heavy metals by sediments through complexation, so the OM → heavy metal path is labeled with "+"; while excessively high concentrations of suspended particulate matter (SS) may inhibit heavy metal fixation by occupying active sites or forming colloidal barrier layers, so the SS → heavy metal path is labeled with "-".

[0038] (4) After completing the initial structure setting, the optimal BN model was selected through parameter learning and structure learning. After iterative optimization by the model validation program, the final BN model was a directed acyclic graph structure, covering the interaction between heavy metals and various environmental factors. Among them, the OM→heavy metal migration path was marked with "+", indicating that the increase of organic matter promoted the adsorption of heavy metals; the SS→heavy metal migration path was marked with "-", indicating that the increase of suspended matter inhibited the fixation of heavy metals. The overall network can show both the causal direction and the positive and negative effects.

[0039] The parameter learning specifically involves: based on the preprocessed data in step (1), conditional probability distribution modeling is performed for both discrete and continuous nodes. For discrete nodes, the conditional probability table (CPT) is calculated by statistically analyzing the frequency of occurrence of each state under different parent node combinations and using a combination of maximum likelihood estimation and Bayesian prior. For continuous nodes, under the assumption that they follow a normal distribution, the distribution parameters such as conditional mean and standard deviation are estimated using the expectation-maximization (EM) algorithm. If the model requires a discrete structure, the continuous variables can be discretized and transformed into finite state variables using the equal interval partitioning method, K-means clustering method, or expert rule segmentation method.

[0040] To further improve the fitting ability of the network structure, an optimization learning mechanism driven by a scoring function is used, based on the initial network structure generated by the causal relationships of variables and their positive and negative effects. The scoring function adopts the Bayesian Information Criterion (BIC) to comprehensively evaluate the model's fitting ability and structural complexity. On this basis, heuristic algorithms such as Greedy Equivalence Search or Tabu Search are used to add, delete, or reverse edges within the allowed edge set space to iteratively optimize the network structure. When the scoring results tend to stabilize or reach the set maximum number of iterations (e.g., 100 times), the structure learning process is terminated, and the network model with the optimal structure is finally selected.

[0041] The completed model undergoes performance evaluation through a model validation procedure, which includes two parts: cross-validation and expert validation. During cross-validation, the dataset is divided into a training set (X%) and a test set (1-X%), with the training set accounting for 80% and the test set accounting for 20%. On the test set, the model's prediction accuracy and goodness of fit are evaluated by calculating metrics such as the root mean square error (RMSE) and the coefficient of determination (R²). Lower RMSE values ​​and higher R² values ​​indicate that the model has strong simulation capabilities.

[0042] The expert verification refers to inviting experts with backgrounds in water environment protection, ecological restoration, or environmental chemistry to review and judge the network structure and output results from aspects such as variable causal logic, rationality of path mechanism, and physical interpretability of model results, in order to help identify potential structural biases or improper path settings.

[0043] In this embodiment, root mean square error (RMSE) and coefficient of determination (R²) were used to evaluate model performance. The RMSE of the adsorption model was 0.05, and the R² was 0.92; the RMSE of the desorption model was 0.06, and the R² was 0.90; the RMSE of the environmental factor competition model was 0.08, and the R² was 0.88. These results indicate that the models have good simulation performance and can accurately predict the adsorption and desorption behavior of heavy metals in lake sediments under different environmental conditions.

[0044] The iterative optimization refers to the following: based on the model validation results, if the prediction error is large, or if expert evaluation points out obvious structural irrationalities, it is necessary to trace back to the key steps in the model construction process and adjust and optimize them layer by layer. At the data level, key missing indicators can be collected, missing value imputation strategies can be optimized, or variable state intervals can be redefined; at the structural level, the direction of node relationships can be adjusted, positive and negative effect labels can be corrected, or a new structural search can be performed within the rules; at the parameter level, the Bayesian prior parameter settings can be optimized or the probability estimation method can be improved. After each round of optimization, the parameter learning and performance evaluation process must be repeated until the model performance meets the set standards.

[0045] (5) Based on the completed BN model, its reasoning mechanism can be used to infer the concentration change trend of heavy metals in the water body and sediment to be tested after inputting specific environmental factor values ​​or states, identify the key factors with the greatest impact and clear direction, and provide a basis for the selection of pollution remediation strategies.

[0046] Specifically, based on the constructed BN model structure and parameters, given the states of some node variables (such as environmental factors), the posterior probability distribution of the states of other target nodes (such as heavy metal speciation concentrations) is calculated. In practice, the heavy metal content in sediments and water bodies are set as target inference variables. Combinations of environmental factor states under different scenarios (such as high OM, low DO, high SS, etc.) are input, and the model outputs the probability or expected value of heavy metal distribution under the corresponding conditions, thus inferring the direction of heavy metal migration and its controlling mechanisms. If, under a certain scenario, the model inference results show a significant increase in the probability of heavy metal release states in water bodies, it can be determined that this scenario (such as low pH, high OM) may induce a heavy metal release process from sediments to water bodies; conversely, if the probability of metal adsorption states in sediments increases, it indicates that the heavy metal migration process from water bodies to sediments is enhanced. Simultaneously, by comparing the differences in model output results under different environmental factor node state changes, the influence of different factors on heavy metal migration and transformation pathways is quantified, which is used to identify the dominant controlling factors and provide a basis for subsequent remediation strategy formulation.

[0047] In this embodiment, analysis using a trained BN model revealed that pH and organic matter content are key factors affecting heavy metal desorption in the lake. A decrease in pH significantly increases heavy metal desorption, while an increase in organic matter content effectively inhibits heavy metal desorption.

[0048] Solution Development: Addressing the aforementioned key factors, a chemical remediation approach is adopted, and the following remediation plan is developed: Add an appropriate amount of alkaline regulator, such as lime, to the lake water to raise the pH value from the current 6.5 to 7.5-8.0. Simultaneously, add biochar to the sediment at a rate of 5 kg per square meter. Biochar can increase the organic matter content of the sediment and enhance its adsorption capacity for heavy metals.

[0049] (6) After the remediation measures are implemented, monitoring data will be continuously collected, the status of key nodes in the BN model will be dynamically updated, and the changes in bioavailable heavy metal content and key factors in water and sediment before and after remediation will be compared and analyzed to verify the remediation effectiveness and optimize the strategy iteratively. Solution Optimization: The BN model was used to predict the effects of the remediation plan. Simulation results showed that after 6 months of implementation, the content of exchangeable heavy metals in the sediments would decrease by 30%–40%, and the content of heavy metals in the overlying water would decrease by 20%–30%. Based on the prediction results, this plan was selected for implementation.

[0050] Repair effect monitoring and feedback: Monitoring: The lake was sampled and monitored in the third and sixth months after the restoration plan was implemented.

[0051] The above description is only a preferred embodiment of the present invention. All equivalent changes and modifications made within the scope of the claims of the present invention should be included in the scope of the present invention.

Claims

1. A method for ecological remediation of heavy metal pollution in river and lake sediments based on the BN model, characterized in that, The method includes the following steps: (1) Collect sediment and overlying water samples from typical estuaries and lakes in the target watershed, determine the concentration of heavy metals and environmental factors affecting the migration and transformation of heavy metals, and preprocess the data. (2) With the goal of reducing the concentration of heavy metals in the water and sediments to be tested, the node variables of the BN model are set based on the preprocessed data obtained in step (1). (3) Combining literature, expert knowledge and sensitivity analysis of existing data, determine the causal relationship between node variables, construct the initial BN network structure, and use "+" and "-" to represent positive and negative effects respectively to label each causal edge, forming a structure diagram containing both causal direction and effect direction information; (4) The optimal structure of the BN model is selected through parameter learning and structure learning. After iterative optimization through the model verification program, the BN model with environmental mechanism support capability is finally constructed. (5) Based on the BN model constructed in step (4), its reasoning mechanism is used to deduce the concentration change trend of heavy metals in the water body and sediment to be tested after inputting specific environmental factor values ​​or states, identify the key environmental factors with the greatest impact and clear direction, and formulate pollution remediation measures accordingly. (6) After the remediation measures are implemented, monitoring data are continuously collected and the status of key nodes of the BN model is dynamically updated. By comparing and analyzing the changes in bioavailable heavy metal content and key environmental factors in the water body and sediment before and after remediation, the remediation effect is verified and the strategy is optimized and iterated.

2. The method according to claim 1, characterized in that, The generalized graph of the BN model with environmental mechanism support capability described in step (4) is a directed acyclic graph consisting of multiple variable nodes. The nodes are connected by directed edges, and "+" and "-" symbols are marked on the edges to indicate the positive and negative effects between variables.

3. The method according to claim 2, characterized in that, The target watershed mentioned in step (1) is an inland river and lake water body.

4. The method according to claim 1, characterized in that, The heavy metals mentioned in step (1) include Cr, Cd, Cu, Zn, As, Hg, Pb and other common heavy metals; wherein, the heavy metal concentration is determined by speciation analysis, including the detection of exchangeable, organically bound and other bioavailable forms.

5. The method according to claim 1, characterized in that, The environmental factors affecting the migration and transformation of heavy metals in step (1) include water environment factors, sediment-related characteristic parameters, interfacial process parameters, and characteristic pollutant indicators; the water environment factors include the pH and temperature of the overlying water body and its total nitrogen, total phosphorus, dissolved organic matter, chloride ions, dissolved oxygen, and soluble fluoride; the sediment-related characteristic parameters include soil particle size, mineral composition, and specific surface area; the interfacial process parameters include microbial activity intensity and sulfide oxidation rate; and the characteristic pollutant indicators include anionic surface activity.

6. The method according to claim 1, characterized in that, The data preprocessing in step (1) includes outlier handling, missing value imputation, and normalization.

7. The method according to claim 1, characterized in that, The node variables in step (2) include heavy metals in the water body to be tested, heavy metals in the sediment, and environmental factors that affect the migration and transformation of heavy metals in the water body and sediment to be tested.

8. The method according to claim 1, characterized in that, In step (3), "+" indicates a positive effect, meaning that increasing the variable will raise the level of the target variable, and "-" indicates a negative effect, meaning that increasing the variable will lower the level of the target variable.

9. The method according to claim 1, characterized in that, The model validation in step (4) includes two parts: cross-validation and expert validation.

10. The method according to claim 1, characterized in that, The remediation measures described in step (5) include one or more combinations of physical remediation, chemical remediation, and biological remediation.

Images