Construction method of denitrification system of iron-sulfur based matrix constructed wetland based on gibbs free energy regulation and machine learning prediction

Abstract

The application discloses a kind of based on Gibbs free energy regulation and machine learning prediction iron sulfur matrix constructed method of denitrification system of constructed wetland, belong to the technical field of ecological environment engineering and sewage biological treatment.The step 1 is based on the characteristics and prior knowledge of constructed wetland, and FeS2 and Fe 0 Mixed matrix;Step 2: based on the mixed matrix of step 1, construct Fe-S substrate configuration driven machine learning model;Step 3: based on the model of step 2, obtain its prediction framework between microbial response and system performance, to realize the intelligentization and visualization of system regulation.The application realizes medium electron release by ΔG° regulation, avoids 100P electron excess and 1P1Z denitrification lag problem.

Description

Technical Field

[0001] This invention belongs to the technical field of ecological environment engineering and biological wastewater treatment, specifically involving a method for constructing an artificial wetland denitrification system based on Gibbs free energy regulation and machine learning prediction. Background Technology

[0002] Constructed wetlands are widely used in wastewater treatment due to their eco-friendly nature and low operating costs. Traditional constructed wetland systems often use inert fillers such as gravel, pumice, or volcanic rock. While these substrates have good physical structure and support, they lack effective electron donors, resulting in insufficient deep removal capacity of nitrogen, especially nitrate nitrogen.

[0003] As research progressed, researchers began introducing fillers with reducing capabilities to enhance the denitrification performance of the system, among which zero-valent iron (Fe) was used. 0 Fe is widely used in constructed wetlands due to its strong initial electron release, low cost, and high chemical stability. 0 During long-term operation, surface passivation can easily hinder electron release, thereby affecting its continuous denitrification capacity; at the same time, it may also cause pH fluctuations in the system, which can have an adverse effect on microbial physiology.

[0004] Pyrite (FeS2), as a naturally stable sulfide ore, can slowly release Fe in an aquatic environment. 2 FeS2, along with SO42-, provides a long-term, stable electron donor source for the system, helping to maintain the reducing properties of the wetland system. FeS2 can provide metabolic substrates for sulfur-oxidizing and iron-reducing microorganisms, activating the TCA cycle and increasing denitrification metabolic flux. However, when FeS2 is used as a single substrate (i.e., 100P configuration), its high Gibbs free energy release characteristic easily leads to excessive electron supply, redox imbalance, rapid depletion of dissolved oxygen (DO), increased microbial metabolic load, and potentially inducing system functional decline. Therefore, some studies have attempted to combine FeS2 with Fe... 0 Compound energy supply system is constructed by blending. Fe 0 While FeS2 releases electrons rapidly but for a short duration, its release is slow but stable over a long period. The synergy between the two can achieve complementarity at different operational stages. However, a systematic understanding of the relationship between their electron release regulation mechanisms and microbial responses is currently lacking, as are predictable design models. Therefore, there is an urgent need to develop a novel Fe–S composite matrix system with tunable configuration, controllable release, a clear response mechanism, and suitability for long-term operation. Summary of the Invention

[0005] This invention provides a method for constructing an artificial wetland denitrification system based on Gibbs free energy regulation and machine learning prediction. By regulating ΔG°, moderate electron release is achieved, avoiding electron excess in 100P and denitrification lag in 1P1Z.

[0006] This invention is achieved through the following technical solution:

[0007] A method for constructing an iron-sulfur-based artificial wetland denitrification system based on Gibbs free energy regulation and machine learning prediction, the method comprising the following steps:

[0008] Step 1: Constructing FeS2 and Fe based on the characteristics and prior knowledge of constructed wetlands 0 Mixed matrix;

[0009] Step 2: Based on the hybrid matrix from Step 1, construct a machine learning model driven by the Fe–S substrate configuration;

[0010] Step 3: Based on the model in Step 2, a predictive framework is obtained between it and microbial response and system performance, which can be used to realize intelligent and visual system regulation.

[0011] Furthermore, step 1 specifically involves clarifying, based on the ΔG° theory, that the 100P configuration suffers from electron excess and increased metabolic burden, while the 1P1Z configuration suffers from insufficient electron supply. Therefore, a 2P1Z configuration with moderate electron release intensity is proposed for FeS2 and Fe... 0 Mixed matrix.

[0012] Furthermore, step 2 specifically includes the following steps:

[0013] Step 2.1: Input variable system and configuration parameterization;

[0014] Step 2.2: Mechanism-guided intermediate variables and interpretable structures;

[0015] Step 2.3: Multi-output supervised model and objective function;

[0016] Step 2.4: Hyperparameter optimization and training process;

[0017] Step 2.5: Verification of the new configuration 2P1Z.

[0018] Furthermore, step 2.1 specifically includes the following steps:

[0019] Step 2.1.2 Joint coding of substrate category and proportion:

[0020] Specifically, category coding involves setting a configuration set. For any configuration Using vectors As a category feature;

[0021] Proportional parameterization: Let r be the volume fraction ratio r = FeS2:Fe 0 r≥0;

[0022] Normalize r into a fractional vector p = (p P ,p Z ),

[0023] p P +p Z =1;

[0024] Step 2.1.2 Characterization of effective thermodynamic driving force specifically involves,

[0025] The intensity of electron donor release under given hydraulic and environmental conditions for different substrates can be determined by the "effective standard Gibbs free energy". For approximate representation; can be As a derived feature to be learned:

[0026]

[0027] Where, φ env Environmental correction factor; It can be obtained by data fitting.

[0028] Furthermore, step 2.2 specifically includes the following steps:

[0029] Step 2.2.1 Ions and metabolic channels:

[0030] To make explicit the causal pathways between thermodynamics, ion release, metabolic regulation, and denitrification performance, the following intermediate variables are introduced and structural equations are established:

[0031]

[0032] η TN =δ0+δ1·Fe 2+ +δ2·TCA act +δ3·DO+ε4

[0033] Where η is the removal rate;

[0034] Step 2.2.2 Electron donor release intensity:

[0035] If a differentiable closed approximation is required during the feature engineering stage, we can let:

[0036]

[0037] And with Fe 2+ ≈a P ·pP ·I e +a Z ·p Z ·I e and As an interpretable prior approximation, the coefficients are then adaptively corrected by the data.

[0038] Furthermore, step 2.3 specifically includes the following steps:

[0039] Step 2.3.1: Model Output:

[0040]

[0041] Step 2.3.2: Multi-objective evaluation function:

[0042] A comprehensive objective combining prediction accuracy, operational stability, and byproduct risk is introduced to minimize:

[0043] J(θ)=w MSE ·L MSE +w stab ·L stab +w risk ·L risk

[0044] in,

[0045]

[0046] Wherein, S_max is the sulfate ecological threshold;

[0047] Step 2.3.3: The embedding of constraints are as follows:

[0048] Feasible region: 0≤p P ≤1,p Z =1-p P r≥0,pH min ≤pH≤pH max ,DO≥0

[0049] Monotonicity constraint:

[0050] Among them, S acc The proxy variable representing the accumulation of sulfides / sulfur-containing byproducts can be approximated by SO42- or online sulfide sensing signals;

[0051] The formula for the integration of Lagrange and soft constraints is:

[0052]

[0053] Furthermore, step 2.4 specifically involves minimizing... Corresponding cross-validation risk; considering that samples with low removal rates are more critical to engineering safety, the loss can be weighted: for y i Lower-weighted samples are given greater weight to improve the model’s sensitivity to “worst-case scenarios”.

[0054] Furthermore, step 2.5 specifically involves, given 2P1Z:

[0055] Construct input

[0056] Use the trained get

[0057] This is the core prediction result of the final output of the entire technical solution.

[0058] A system for constructing an iron-sulfur-based constructed wetland nitrogen removal system based on Gibbs free energy regulation and machine learning prediction is disclosed. The system utilizes the aforementioned method for constructing an iron-sulfur-based constructed wetland nitrogen removal system based on Gibbs free energy regulation and machine learning prediction. The system comprises:

[0059] Hybrid matrix construction module: Constructing FeS2 and Fe based on the characteristics and prior knowledge of constructed wetlands. 0 Mixed matrix;

[0060] Machine learning model building module: Based on the hybrid matrix built by the hybrid matrix building module, construct a machine learning model driven by Fe–S substrate configuration;

[0061] Prediction framework: Based on the model built by the machine learning model building module, a prediction framework is obtained between the model and the microbial response and system performance, which is used to realize the intelligent and visual regulation of the system.

[0062] An application of the above-mentioned method for constructing an iron-sulfur-based artificial wetland denitrification system based on Gibbs free energy regulation and machine learning prediction is applied to an artificial wetland that balances reaction rate and stability to regulate electron release intensity.

[0063] The beneficial effects of this invention are:

[0064] This invention is based on FeS2 and Fe 0 A dual-channel electronic power supply mechanism is collaboratively constructed, proposing 2P1Z(FeS2:Fe 0 =2:1) ​​configuration, by controlling the electron release intensity through moderate Gibbs free energy, avoids the dual drawbacks of excess electrons in 100P and insufficient electrons in 1P1Z, and constructs a highly efficient denitrification system that takes into account both reaction rate and stability.

[0065] This invention regulates the intensity of electron release to support microbial metabolism.

[0066] The experimental results of this invention show that 2P1Z significantly enhances Fe 2+ It stabilizes SO42- release, maintains DO balance, significantly activates the TCA cycle, and upregulates the expression of key Fe / N transport genes and denitrification functional genes. Attached Figure Description

[0067] Figure 1 This is a diagram of the experimental setup for the continuous subsurface flow constructed wetland system of the present invention.

[0068] Figure 2 The pollutant removal efficiency of the 2P1Z constructed wetland of the present invention is shown in (a) for the COD removal rate of the three iron matrices over 200 days, (b) for the TN removal rate of the three iron matrices over 200 days, (c) for the NO3--N removal rate of the three iron matrices over 200 days, and (d) for the NH4+-N removal rate of the three iron matrices over 200 days.

[0069] Figure 3 The present invention is 2P1Z Fe 2+ The dynamics of SO42- release are shown, where (a) represents the change in ferrous iron release over 200 days for the three iron matrices, and (b) represents the change in sulfate release over 200 days for the three iron matrices.

[0070] Figure 4 This invention presents a comparison between predicted and measured nitrogen removal performance based on the XGBoost model, showing (a) the predicted and measured COD removal efficiency in the 2P1Z system; (b) the predicted and measured TN removal time series in the 2P1Z system; and (c) the predicted and measured NO3 levels. - -N removal comparison; (d) Predicted and measured NH4 values ​​during operation + -N removes the effect. Detailed Implementation

[0071] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of this application with unnecessary detail.

[0072] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0073] It should also be understood that the terminology used in this application specification is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this application specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0074] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0075] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0076] Implementation Method 1

[0077] This embodiment provides a method for constructing an iron-sulfur-based artificial wetland denitrification system based on Gibbs free energy regulation and machine learning prediction. The construction method includes the following steps:

[0078] Step 1: Constructing FeS2 and Fe based on the characteristics and prior knowledge of constructed wetlands 0 Mixed matrix;

[0079] Step 2: Based on the hybrid matrix from Step 1, construct a machine learning model driven by the Fe–S substrate configuration;

[0080] Step 3: Based on the model in Step 2, a predictive framework is obtained between it and microbial response and system performance, which can be used to realize intelligent and visual system regulation.

[0081] Furthermore, step 1 specifically involves clarifying, based on the ΔG° theory, that the 100P configuration suffers from electron excess and increased metabolic burden, while the 1P1Z configuration suffers from insufficient electron supply. Therefore, a 2P1Z configuration with moderate electron release intensity is proposed for FeS2 and Fe... 0 Mixed matrix.

[0082] Furthermore, step 2 specifically includes the following steps:

[0083] Step 2.1: Input variable system and configuration parameterization;

[0084] Step 2.2: Mechanism-guided intermediate variables and interpretable structures;

[0085] Step 2.3: Multi-output supervised model and objective function;

[0086] Step 2.4: Hyperparameter optimization and training process;

[0087] Step 2.5: Verification of the new configuration 2P1Z.

[0088] Furthermore, step 2.1 specifically includes the following steps:

[0089] Step 2.1.2 Joint coding of substrate category and proportion:

[0090] Category encoding (one-hot or embedding) specifically involves setting a configuration set. For any configuration Using vectors As a categorical feature (or using a low-dimensional embedding vector);

[0091] Proportional parameterization: Let r be the volume fraction ratio r = FeS2:Fe 0 r≥0 (when it is pure FeS2, we consider r→+∞);

[0092] Normalize r into a fractional vector p = (p P ,p Z ),

[0093] p P +p Z =1;

[0094] Step 2.1.2 Characterization of effective thermodynamic drive (which can be used as a derived input) specifically involves:

[0095] The intensity of electron donor release under given hydraulic and environmental conditions for different substrates can be determined by the "effective standard Gibbs free energy". For approximate representation; can be As a derived feature to be learned:

[0096]

[0097] Where, φ env Environmental correction factors (such as pH, DO, temperature, ionic strength, etc.); It can be obtained by data fitting.

[0098] Furthermore, step 2.2 specifically includes the following steps:

[0099] Step 2.2.1 Ions and metabolic channels:

[0100] To make explicit the causal pathways between thermodynamics, ion release, metabolic regulation, and denitrification performance, the following intermediate variables are introduced and structural equations are established:

[0101]

[0102] η TN =δ0+δ1·Fe 2+ +δ2·TCA act +δ3·DO+ε4

[0103] Where η is the removal rate, and the formulas for other pollutants can be constructed similarly;

[0104] Step 2.2.2 Electron donor release intensity:

[0105] If a differentiable closed approximation is required during the feature engineering stage, we can let:

[0106]

[0107] And with Fe 2+ ≈a P ·p P ·I e +a Z ·p Z ·I e and As an interpretable prior approximation, the coefficients are then adaptively corrected by the data.

[0108] Furthermore, step 2.3 specifically includes the following steps:

[0109] Step 2.3.1: Model Output:

[0110]

[0111] Step 2.3.2: Multi-objective evaluation function:

[0112] A comprehensive objective combining prediction accuracy, operational stability, and byproduct risk is introduced to minimize:

[0113] J(θ)=w MSE ·L MSE +w stab ·L stab +w risk ·L risk

[0114] in,

[0115] Variance is calculated from the first difference of the time series to suppress fluctuations that cause sudden changes in brightness.

[0116]

[0117] Wherein, S_max is the sulfate ecological threshold;

[0118] Step 2.3.3: The embedding of constraints are as follows:

[0119] Feasible region: 0≤p P ≤1,p Z =1-p P r≥0,pH min ≤pH≤pH max ,DO≥0

[0120] Monotonicity constraint:

[0121] Among them, S acc The proxy variable representing the accumulation of sulfides / sulfur-containing byproducts can be approximated by SO42- or online sulfide sensing signals;

[0122] The formula for the integration of Lagrange and soft constraints is:

[0123]

[0124] Furthermore, step 2.4 specifically involves minimizing... Corresponding cross-validation risk; considering that samples with low removal rates are more critical to engineering safety, the loss can be weighted: for y i Lower-weighted samples are given greater weight to improve the model’s sensitivity to “worst-case scenarios”.

[0125] Furthermore, step 2.5 specifically involves, given 2P1Z:

[0126] Construct input

[0127] Use the trained get

[0128] This is the core prediction result of the final output of the entire technical solution.

[0129] It is a pre-trained machine learning model Given a specific input x *The resulting predicted output vector, according to the documentation, contains predicted values ​​for key performance indicators of the constructed wetland system.

[0130]

[0131] This represents the model's adoption of a new "electron release intensity enhanced" (2P1Z) configuration, and the iron-sulfur matrix (FeS2:Fe) 0 The final predicted results of the denitrification performance (total nitrogen, nitrate nitrogen, and ammonia nitrogen removal rate) and organic matter removal performance (COD removal rate) of an constructed wetland system with a 2:1 ratio under specific operating conditions.

[0132]

[0133]

[0134] This invention constructs a continuous subsurface flow artificial wetland, using pyrite and zero-valent iron (uniformly mixed, volume ratio 1:1) as the main matrix layer, with 60 plants / m². 2 Sweet flag was planted at a specific density. Subsequently, a composite substrate consisting of pure pyrite matrix and pyrite with zero-valent iron in a 1:1 ratio was used as a control. The three constructed wetland systems were placed in a greenhouse and operated stably for 60 days, with pollutant removal efficiency monitored.

[0135] Example: Preparation of pyrite and zero-valent iron

[0136] Pyrite (FeS2) needs to be soaked in 0.1 mol / L hydrochloric acid (HCl) for 30 minutes before use to remove surface oxides and impurities. It should then be rinsed with deionized water until the pH of the washing solution is close to neutral. After filtration, it should be dried in a 60℃ oven for 12 hours before use. Zero-valent iron (Fe) 0 First, ultrasonically clean it with deionized water for 15 minutes to remove surface particulate impurities and oxide layer, and then seal and store it in an N2 atmosphere to prevent it from being oxidized by air.

[0137] The three substrate configurations are as follows:

[0138] 100P system: Uses 100% pure FeS2 as the functional matrix, directly packed into the reactor without doping with other electron donors. This configuration represents high reduction strength conditions. 1P1Z system: Combines FeS2 with Fe... 0 Mix uniformly at a 1:1 volume ratio using a drum mixer with slow rotation for 30 minutes to ensure uniform particle distribution. This configuration is used to simulate a medium-intensity electronic power supply system. 2P1Z system: According to FeS2:Fe... 0 The filler is mixed in a 2:1 volume ratio using the same method as above to obtain a composite filler with enhanced reducing properties, which combines high reactivity with a certain buffering capacity.

[0139] During the experiment, simulated domestic sewage was continuously injected into each reactor. The concentrations of the main pollutants were: COD 280 mg / L and NH4+. + -N 35mg / L, NO3 - -N 15mg / L, NO2 - -N 5 mg / L, TN 55 mg / L. Hydraulic retention time (HRT) was 5 days, with the liquid level controlled 5 cm above the substrate layer, and gravity outflow achieved through bottom drainage. The first 14 days were an acclimatization period, with intermittent aeration for 12 h / 12 ​​h to promote microbial biofilm adhesion. From day 15 onwards, continuous anaerobic operation was switched to continue until day 200. After system stabilization (from day 60 onwards), effluent samples were collected every 5 days at sampling depths of 10 cm and 20 cm in the substrate layer. Immediately after sampling, the water samples were filtered (using a 0.45 μm membrane) and centrifuged at 12,000 rpm to determine water quality indicators including COD, TN, and NH4+. + -N and NO3 - -N, the method is based on the national standard "Methods for Monitoring and Analysis of Water and Wastewater" (GB / T 5750). Fe 2+ The determination of SO4 was performed using the o-phenanthroline spectrophotometric method (510 nm). 2- The results were determined by barium chromate colorimetry (420 nm), with all indicators analyzed in triplicate to ensure data accuracy. To investigate the system's thermodynamic characteristics and ion release behavior, the theoretical energy release intensity under different matrix configurations was calculated using the standard Gibbs free energy change (ΔG°) formula. The standard enthalpy changes and entropy values ​​of relevant substances were obtained from the NIST database, and the calculation results were used to evaluate electron release capacity and reaction driving force. The ΔG° values ​​were: 100P = –2427.0 kJ / mol, 1P1Z = –1102.7 kJ / mol, and 2P1Z = –1813.4 kJ / mol. Simultaneously, this invention introduces an XGBoost machine learning model for predicting and inverting pollutant removal rates. The model takes the Fe–S matrix type as input and outputs four pollutants (COD, TN, NO3). - -N, NH4 + -N) Removal efficiency. Training data comes from measured data of the 100P and 1P1Z systems over a period of 60–200 days. The model was subjected to five-fold cross-validation, and R... 2 All values ​​are greater than 0.85, with a prediction error of less than 5%. These values ​​were subsequently used to predict the performance of the 2P1Z configuration and verified using measured data. The following examples are some specific implementation and application cases of this invention in practical applications, but are not limited to these.

[0140] Application example: Constructed wetland using pyrite and zero-valent iron in a 2:1 ratio as filler.

[0141] This experiment constructed two sets of continuous subsurface flow constructed wetlands. The constructed wetland system used pyrite and zero-valent iron as the main substrate layer (uniformly mixed, volume ratio 2:1). Sweet flag (Acorus calamus L.) was used as the test plant, with a planting density of 60 plants / m². 2 The biomass is 1.0 kg fresh weight / m³. 2 After the constructed wetland system is completed, simulated domestic sewage will be introduced from the top of the wetland using a peristaltic pump, with a hydraulic retention time of 5 days. Additionally, limestone should be added to the substrate to prevent sedimentation and clogging.

[0142] Compared to 1P1Z and 100P, 2P1Z significantly improved the pollutant removal efficiency of the constructed wetland system at the end of a 200-day operating period. Specifically, the 2P1Z group showed improvements in COD, TN, and NH4+. + -N and NO3 - The nitrogen removal rates were 90.06% ± 3.45%, 89.46% ± 4.21%, 92.95% ± 4.23%, and 93.51% ± 3.87%, respectively, combining the advantages of high stability and high nitrogen removal rate of the 1P1Z and 100P groups.

[0143] Implementation Method 2

[0144] This embodiment provides a system for constructing an iron-sulfur-based constructed wetland nitrogen removal system based on Gibbs free energy regulation and machine learning prediction. The system uses the method described in Embodiment 1 for constructing an iron-sulfur-based constructed wetland nitrogen removal system based on Gibbs free energy regulation and machine learning prediction. The system includes:

[0145] Hybrid matrix construction module: Constructing FeS2 and Fe based on the characteristics and prior knowledge of constructed wetlands. 0 Mixed matrix;

[0146] Machine learning model building module: Based on the hybrid matrix built by the hybrid matrix building module, construct a machine learning model driven by Fe–S substrate configuration;

[0147] Prediction framework: Based on the model built by the machine learning model building module, a prediction framework is obtained between the model and the microbial response and system performance, which is used to realize the intelligent and visual regulation of the system.

[0148] Implementation Method 3

[0149] This embodiment provides an application of a method for constructing an iron-sulfur-based artificial wetland denitrification system based on Gibbs free energy regulation and machine learning prediction, as described in Embodiment 1. It is applied to an artificial wetland that is a highly efficient denitrification system that balances reaction rate and stability, in order to regulate the intensity of electron release.

Claims

1. A method for constructing an artificial wetland denitrification system based on Gibbs free energy regulation and machine learning prediction, characterized in that, The construction method includes the following steps: Step 1: Constructing FeS2 and Fe based on the characteristics and prior knowledge of constructed wetlands 0 Mixed matrix; Step 2: Based on the hybrid matrix from Step 1, construct a machine learning model driven by the Fe–S substrate configuration; Step 3: Based on the model in Step 2, obtain the predictive framework between it and microbial response and system performance, so as to realize the intelligent and visual regulation of the system; Step 2 specifically includes the following steps: Step 2.1: Input variable system and configuration parameterization; Step 2.2: Mechanism-guided intermediate variables and interpretable structures; Step 2.3: Multi-output supervised model and objective function; Step 2.4: Hyperparameter optimization and training process; Step 2.5: Verification of the new configuration 2P1Z; Step 2.2 specifically includes the following steps: Step 2.2.1 Ions and Metabolic Channels: To make explicit the causal pathways between thermodynamics, ion release, metabolic regulation, and denitrification performance, the following intermediate variables are introduced and structural equations are established: in, For removal rate, These are the regression coefficients of the structural equation model. These are the regression coefficients of the structural equation model. These are the regression coefficients of the structural equation model. These are the regression coefficients of the structural equation model. For the perturbation term / residual of the equation, For the perturbation term / residual of the equation, For the perturbation term / residual of the equation, For the perturbation term / residual of the equation, These are the regression coefficients of the structural equation model. These are the regression coefficients of the structural equation model. These are the regression coefficients of the structural equation model. These are the regression coefficients of the structural equation model. These are the regression coefficients of the structural equation model. These are the regression coefficients of the structural equation model. The degree of activation of metabolic channels; Step 2.2.2 Electron donor release intensity: If a differentiable closed approximation is required during the feature engineering stage, we can let: in, This is an approximate indicator of the intensity of electron donor release. For environment correction functions, pH level For dissolved oxygen, For temperature; and with and As an interpretable prior approximation, the coefficients are then adaptively corrected by the data. in, for Fraction, for Fraction.

2. The method according to claim 1, characterized in that, Step 1 specifically involves clarifying, based on the ΔG° theory, that the 100P configuration suffers from electron excess and increased metabolic burden, while the 1P1Z configuration suffers from insufficient electron supply. Therefore, a 2P1Z configuration with moderate electron release intensity is proposed for FeS2 and Fe... 0 Mixed matrix.

3. The method according to claim 1, characterized in that, Step 2.1 specifically includes the following steps: Step 2.1.2 Joint coding of substrate category and proportion: Specifically, category coding involves setting a configuration set. ={100P,1P1Z,2P1Z,…}, for any configuration c∈ Using vector e(c)∈{0,1}^{| |} serves as a category feature; Proportional parameterization: Let r be the volume fraction ratio ; Normalize r into a fractional vector , ； in, for Fraction, for Fraction; Step 2.1.2 Characterization of effective thermodynamic driving force specifically involves, The intensity of electron donor release under given hydraulic and environmental conditions for different substrates can be determined by the "effective standard Gibbs free energy". For approximate representation; can be As a derived feature to be learned: in, Environmental correction factor; It can be obtained by data fitting. c For substrate configuration, Let be the fractional vector of the volume fraction ratio.

4. The method according to claim 1, characterized in that, Step 2.3 specifically includes the following steps: Step 2.3.1: Model Output: in, Input to the model, This is the true removal rate vector. To predict the output, For model parameters, COD pollutant removal rate, TN pollutant removal rate, NO3 pollutant removal rate The removal rate of NH4 pollutants; Step 2.3.2: Multi-objective evaluation function: A comprehensive objective combining prediction accuracy, operational stability, and byproduct risk is introduced to minimize: in, , , Weights for MSE loss. For mean square error loss, For stability loss, For stability loss weights, For by-product risk loss weighting, Risk loss due to byproducts; Among them, S max The sulfate ecological threshold, For the sample size, For the first i Predicted sulfate concentration for a single sample or time point. i For sample index; Step 2.3.3: The embedding of constraints are as follows: Feasible domain: Monotonicity constraint: in, A proxy variable representing the accumulation of sulfides / sulfur-containing byproducts can be represented by SO4². - Or the online sulfide sensing signal is similar; The formula for the integration of Lagrange and soft constraints is: in, For soft constraint penalty weights, For mathematical expectation, For soft constraint penalty weights, This is the original multi-objective evaluation function.

5. The method according to claim 1, characterized in that, Step 2.4 specifically involves minimizing... Corresponding cross-validation risk; considering that samples with low removal rates are more critical to engineering safety, loss weighting can be applied: for Lower-weighted samples are given greater weight to improve the model’s sensitivity to “worst-case scenarios”.

6. The method according to claim 1, characterized in that, Specifically, step 2.5 involves giving... Construct input , Use the trained get ; This is the core prediction result of the final output of the entire technical solution.

7. A system for constructing an artificial wetland nitrogen removal system based on Gibbs free energy regulation and machine learning prediction, characterized in that, The system uses the method for constructing an iron-sulfur-based artificial wetland denitrification system based on Gibbs free energy regulation and machine learning prediction as described in any one of claims 1-6, and the system comprises: Hybrid matrix construction module: Constructing FeS2 and Fe based on the characteristics and prior knowledge of constructed wetlands. 0 Mixed matrix; Machine learning model building module: Based on the hybrid matrix built by the hybrid matrix building module, construct a machine learning model driven by Fe–S substrate configuration; Prediction framework: Based on the model built by the machine learning model building module, a prediction framework is obtained between the model and the microbial response and system performance, which is used to realize the intelligent and visual regulation of the system.

8. An application of the method for constructing an artificial wetland denitrification system based on Gibbs free energy regulation and machine learning prediction as described in any one of claims 1-6, characterized in that, Artificial wetlands used in efficient denitrification systems that balance reaction rate and stability are used to regulate the intensity of electron release.

Images