Mechanism-data hybrid driven modeling and control method for wastewater biochemical treatment process

Abstract

This invention discloses a mechanism-data hybrid-driven method for modeling and controlling wastewater biochemical treatment processes, comprising the following steps: establishing a mechanism model for the target pollutant; screening key kinetic parameters with high sensitivity and high volatility; using experimental datasets to invert the key kinetic parameter dataset based on the mechanism model; constructing a prediction model for the key kinetic parameters using a data-driven algorithm; coupling the prediction model and the mechanism model into a hybrid prediction model, and inputting real-time influent and operating conditions to dynamically predict the concentrations of the target pollutant and its products; based on the prediction results, solving a multi-objective optimization problem with target pollutant removal and product resource recovery as optimization objectives, and obtaining and executing the optimal control parameters. This invention reduces prediction fluctuation bias and improves prediction robustness by constraining the main mechanism framework, and improves prediction accuracy by dynamically fitting highly sensitive and highly volatile parameters using a data-driven algorithm, thereby achieving synergistic optimization of pollutant removal and resource recovery.

Description

Technical Field

[0001] This invention belongs to the field of environmental wastewater data processing technology. Specifically, it relates to a mechanism-data hybrid driven method for modeling and controlling wastewater biochemical treatment processes. Background Technology

[0002] With the development of urbanization and industrial activities, wastewater treatment processes face challenges such as increased fluctuations in water quality and quantity, and complex and variable operating conditions, exhibiting characteristics of strong nonlinearity, time-varying nature, and high uncertainty. Constructing computational models that can accurately characterize the dynamic behavior of the treatment process and provide reliable support for operational control and decision-making has become a key technological foundation for the optimization and intelligent control of modern wastewater treatment systems. Currently, wastewater treatment process modeling mainly employs two methods: mechanistic models and data-driven models.

[0003] Mechanistic models provide a structured description of treatment processes through biochemical reaction kinetics and mass transfer relationships, offering advantages such as clear physical meaning and strong interpretability. However, mechanistic models typically rely on a large number of kinetic parameters that are difficult to measure directly; these parameters are often obtained through inversion under specific experimental conditions or set empirically. In actual dynamic operation, influent water quality, quantity, and operating conditions change frequently, making it difficult for fixed parameter assumptions to reflect the real-time response characteristics of the system, thus reducing the model's predictive accuracy and generalization ability.

[0004] Data-driven approaches do not rely on prior mechanisms and can learn the nonlinear characteristics of a system from historical data. However, they still have limitations: on the one hand, wastewater quality and quantity data still rely on manual sampling and laboratory analysis, and the large amount of high-quality data required for data-driven approaches is costly and lacks completeness, affecting the model training effect; on the other hand, pure data-driven models lack mechanistic constraints, making prediction results difficult to interpret and verify, and their reliability is insufficient in high-risk operational decision-making scenarios.

[0005] To combine the advantages of both types of models, existing technologies have developed several hybrid modeling approaches. First, data-driven models are trained by fusing the output of mechanistic models with experimental data; however, this does not address the problem of fixed parameters and difficulty in dynamic response inherent in mechanistic models. Second, data generated from mechanistic models is used to train surrogate models. While this improves computational speed, it sacrifices the interpretability of the mechanism, and the performance of the surrogate model is limited by the coverage of the training data.

[0006] In summary, existing technologies, whether single mechanistic models, data-driven models, or hybrid methods such as result-level fusion or surrogate models, struggle to simultaneously meet the engineering requirements of high prediction accuracy, strong dynamic adaptability, good physical interpretability, and low dependence on data scale in complex and ever-changing wastewater treatment operating environments. Particularly when facing real-time optimization control needs for multiple objectives (such as the synergistic effect of efficient pollutant removal and resource recovery), there is a lack of modeling and control technology systems capable of continuously and accurately characterizing process dynamics and supporting reliable control decisions. Summary of the Invention

[0007] The primary objective of this invention is to overcome the shortcomings and deficiencies of existing technologies and provide a mechanism-data hybrid-driven modeling and control method for wastewater biochemical treatment processes. This method can improve the accuracy of dynamic prediction of the model while maintaining the interpretability of the mechanism, and achieve synergistic optimization of pollutant removal and resource recovery processes.

[0008] The second objective of this invention is to provide a mechanism-data hybrid driven modeling and control system for wastewater biochemical treatment processes.

[0009] The objective of this invention is achieved through the following technical solution: a mechanism-data hybrid-driven method for modeling and controlling wastewater biochemical treatment processes, comprising the following steps:

[0010] S1. Establish and calibrate a biotransformation mechanism model of the target pollutant to obtain the calibrated mechanism model and the range of kinetic parameter changes after calibration;

[0011] S2. Perform sensitivity analysis and coefficient of variation analysis on each dynamic parameter of the mechanism model to screen out key dynamic parameters;

[0012] S3. Obtain the experimental dataset of the biotransformation process of the target pollutant, and perform parameter inversion based on the mechanism model to obtain the corresponding key kinetic parameter dataset;

[0013] S4. Using the experimental dataset as the input dataset and the key dynamic parameter dataset as the output dataset, a prediction model for the key dynamic parameters is constructed using a data-driven algorithm.

[0014] S5. Couple the prediction model and the mechanism model online to construct a hybrid prediction model; when the hybrid prediction model is running, based on the real-time water inflow conditions and operating conditions, the prediction model dynamically updates the values ​​of key dynamic parameters in the mechanism model.

[0015] S6. Input the real-time influent and operating conditions into the mixing prediction model to dynamically predict the concentration of the target pollutants and their conversion products;

[0016] S7. Based on the prediction results of the hybrid prediction model and the preset control target, construct and solve a multi-objective optimization problem with the target pollutant removal rate and the target product resource recovery rate as optimization objectives, obtain the optimal control parameters, and perform real-time control.

[0017] Preferably, step S1 specifically includes:

[0018] S11. Based on the main biological reaction pathways and transformation relationships of the target pollutants, a mechanistic model framework for microbial dynamics is constructed, which includes reaction equations, biochemical stoichiometry matrices, and microbial dynamics equations.

[0019] S12. Through laboratory sequential batch experiments, collect concentration data of target pollutants and their transformation products over time under different operating conditions to form a calibration dataset;

[0020] S13. Using the calibration dataset, fit and calibrate the dynamic parameters in the mechanism model framework to obtain the calibrated mechanism model and the range of variation of the calibrated dynamic parameters.

[0021] Preferably, step S2 specifically includes:

[0022] S21. Based on the range of variation of the dynamic parameters, calculate the total sensitivity index of each dynamic parameter to evaluate the degree of influence of each dynamic parameter on the output of the mechanism model;

[0023] S22. Based on multiple sets of values ​​of the dynamic parameters obtained under different experimental conditions within the range of variation of the dynamic parameters, calculate their coefficient of variation to assess the degree of fluctuation of each dynamic parameter with operating conditions.

[0024] S23. Based on the comprehensive evaluation results of the total sensitivity index and coefficient of variation of each dynamic parameter, key dynamic parameters are selected.

[0025] Preferably, step S3 specifically includes:

[0026] S31. Obtain experimental data related to the transformation of the target pollutant from publicly published literature, and construct an experimental dataset after screening and standardization.

[0027] S32. Input the inflow conditions from the experimental dataset into the mechanism model, and output the observation results that match the corresponding experiment through parameter calibration to obtain the corresponding key dynamic parameter dataset.

[0028] Preferably, step S4 specifically includes:

[0029] S41. Based on data-driven algorithms, construct an initial key dynamic parameter prediction model;

[0030] S42. Using the influent conditions and operating conditions of the experimental dataset as input variables and the key dynamic parameters in the corresponding key dynamic parameter dataset as output variables, establish an input dataset and an output dataset, and divide them into a training set and a validation set according to the proportion.

[0031] S43. Use the training set to train the prediction model of step S41, and use the validation set to optimize the model hyperparameters until the prediction error of the model on the validation set reaches the preset condition.

[0032] S44. Retrain the optimized prediction model using the complete input and output datasets to obtain the final prediction model for the key dynamic parameters.

[0033] Preferably, in step S4, the data-driven algorithm is a data-driven algorithm based on categorical feature enhancement.

[0034] Preferably, step S7 specifically includes:

[0035] S71. Perform interpretability analysis on the input variables in the hybrid prediction model, identify the degree of influence of each input variable on the model prediction results, and identify key control parameters;

[0036] S72. Construct a dual-objective optimization model with the optimization objectives of maximizing the removal efficiency of target pollutants and the resource recovery rate of target products.

[0037] S73. Based on the feasible range of the key control parameters in step S71, the concentration of the target pollutant and its transformation products predicted by the hybrid prediction model, and the preset control target, solve the dual-objective optimization model to obtain the optimal combination of control parameters.

[0038] Preferably, the target pollutant removal efficiency and the target product resource recovery rate are calculated according to the following formulas:

[0039] , formula (1),

[0040] , formula (2),

[0041] in, The target pollutant removal efficiency (%). The target product resource recovery rate (%). The target pollutant concentration (mg / L) in the influent. The concentration of the target pollutant in the effluent (mg / L) is given. ν represents the product concentration (mg / L), and v is the control parameter.

[0042] Preferably, the target pollutant is a sulfide, the conversion product includes elemental sulfur, and the multi-objective optimization is to simultaneously maximize the sulfide removal rate and the elemental sulfur recovery rate.

[0043] Mechanism-data hybrid driven modeling and control systems for wastewater biochemical treatment processes include:

[0044] The mechanism model building module is used to establish and calibrate the biotransformation mechanism model of the target pollutant, and obtain the calibrated mechanism model and the range of kinetic parameter changes after calibration.

[0045] The key parameter screening module is used to perform sensitivity analysis and coefficient of variation analysis on each dynamic parameter of the mechanism model to screen out key dynamic parameters.

[0046] The prediction model dataset acquisition module is used to acquire experimental datasets of the biotransformation process of the target pollutant and perform parameter inversion based on the mechanism model to obtain the corresponding key kinetic parameter datasets.

[0047] The prediction model building module is used to construct a prediction model for the key dynamic parameters using the experimental dataset as the input dataset and the key dynamic parameter dataset as the output dataset, and by employing a data-driven algorithm.

[0048] A hybrid prediction model construction module is used to couple the prediction model and the mechanism model online to construct a hybrid prediction model. When the hybrid prediction model is running, the prediction model dynamically updates the values ​​of key dynamic parameters in the mechanism model based on real-time water inflow conditions and operating conditions.

[0049] The dynamic prediction module is used to input real-time influent and operating conditions into the mixing prediction model to dynamically predict the concentration of target pollutants and their conversion products.

[0050] The optimization and control module is used to construct and solve a multi-objective optimization problem with the target pollutant removal rate and target product resource recovery rate as optimization objectives based on the prediction results of the hybrid prediction model and the preset control target, to obtain the optimal control parameters and perform real-time control.

[0051] The present invention has the following advantages and effects compared with the prior art:

[0052] (1) This invention provides a mechanism-data hybrid-driven modeling and control method for wastewater biochemical treatment processes. Using a mechanism model of the target pollutant transformation process as the main structure, and targeting key kinetic parameters identified through sensitivity analysis and parameter fluctuation analysis (coefficient of variation analysis), a data-driven prediction model is introduced. This achieves the fusion modeling of the mechanism model and the data-driven model at the parameter level, resulting in a hybrid prediction model. On the one hand, this method dynamically enhances the key kinetic parameters with high sensitivity and high fluctuation in the mechanism model, realizing continuous prediction and dynamic characterization of target pollutants and their transformation products during wastewater treatment, and improving the prediction accuracy of the model under complex dynamic conditions (fluctuations in influent water quality and changes in operating conditions). On the other hand, this method constrains the results of the hybrid prediction model within a reasonable range of kinetics through the framework of the mechanism model, reducing the additional fluctuations introduced by the data-driven method and improving the prediction robustness of the model under complex dynamic conditions (fluctuations in influent water quality and changes in operating conditions).

[0053] (2) By combining sensitivity analysis and parameter fluctuation analysis, this invention jointly evaluates the dynamic parameters in the mechanism model, identifies key dynamic parameters that have a significant impact on the model output and exhibit obvious fluctuation characteristics with changes in operating conditions, thereby minimizing the complexity and data requirements of data-driven modeling. While improving model performance, it effectively ensures interpretability and reduces dependence on large amounts of data.

[0054] (3) This invention constructs a multi-objective optimization model with the target pollutant removal rate and the target product resource recovery rate as optimization objectives, and generates the optimal control parameters based on the output of the hybrid prediction model. This achieves the synergistic optimization of the pollutant removal and resource recovery process, and achieves the effect of improving treatment efficiency while taking into account resource utilization. It overcomes the problem that it is difficult to achieve efficient removal and high-value recovery at the same time in existing methods, and provides decision support for the intelligent and refined operation of sewage treatment plants that has both environmental and resource benefits. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of the mechanism-data hybrid driven wastewater biochemical treatment process modeling and control method of the present invention.

[0056] Figure 2 This is a prediction effect diagram of high-sulfur wastewater in industrial and mining wastewater using the mechanism-data hybrid driven wastewater biochemical treatment process modeling and control method in Embodiment 1 of the present invention.

[0057] Figure 3 This diagram illustrates the control effect of high-sulfur wastewater in industrial and mining wastewater using the mechanism-data hybrid driven wastewater biochemical treatment process modeling and control method in Embodiment 1 of the present invention. Detailed Implementation

[0058] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0059] Example 1

[0060] This invention provides a mechanism-data hybrid driven method for modeling and controlling wastewater biochemical treatment processes, comprising the following steps:

[0061] S1. Establish and calibrate a biotransformation mechanism model of the target pollutant to obtain the calibrated mechanism model and the range of kinetic parameter changes after calibration;

[0062] S2. Perform sensitivity analysis and coefficient of variation analysis on each dynamic parameter of the mechanism model to screen out key dynamic parameters;

[0063] S3. Obtain the experimental dataset of the biotransformation process of the target pollutant, and perform parameter inversion based on the mechanism model to obtain the corresponding key kinetic parameter dataset;

[0064] S4. Using the experimental dataset as the input dataset and the key dynamic parameter dataset as the output dataset, a prediction model for the key dynamic parameters is constructed using a data-driven algorithm.

[0065] S5. Couple the prediction model and the mechanism model online to construct a hybrid prediction model; when the hybrid prediction model is running, based on the real-time water inflow conditions and operating conditions, the prediction model dynamically updates the values ​​of key dynamic parameters in the mechanism model.

[0066] S6. Input the real-time influent and operating conditions into the mixing prediction model to dynamically predict the concentration of the target pollutants and their conversion products;

[0067] S7. Based on the prediction results of the hybrid prediction model and the preset control target, construct and solve a multi-objective optimization problem with the target pollutant removal rate and the target product resource recovery rate as optimization objectives, obtain the optimal control parameters, and perform real-time control.

[0068] Specifically, the method of this invention uses a microbial kinetic mechanism model as its core. Through sensitivity analysis and coefficient of variation analysis, it identifies condition-sensitive key kinetic parameters within the mechanism model and utilizes data-driven algorithms to accurately predict these key kinetic parameters, thereby constructing a hybrid prediction model that combines the mechanism model with a data-driven prediction model. This invention enhances the predictive ability of the mechanism model for wastewater treatment processes under highly fluctuating and nonlinear operating conditions. It overcomes the problems of insufficient simulation accuracy of traditional static mechanism models under dynamic wastewater conditions and poor interpretability and large fluctuations in prediction results of single data-driven models under actual operating conditions. By introducing data-driven modeling only for a subset of kinetic parameters highly sensitive to operating conditions, it significantly reduces the model's dependence on large-scale, high-quality data while maintaining the integrity of the mechanism structure, and enhances the interpretability of the model results, thereby improving the reliability of the model for process analysis and operational guidance in engineering applications.

[0069] Those skilled in the art will understand that the method described in this invention has broad applicability. Its core lies in the modeling concept and implementation process of parameter-level fusion, rather than being specific to a particular pollutant. For other pollutants in wastewater treatment that can be removed through microbial transformation processes, such as nitrogen-containing compounds (ammonia nitrogen, nitrate nitrogen, etc.), phosphorus-containing compounds, or various organic pollutants, as long as a corresponding microbial kinetic mechanism model is established based on its specific biotransformation pathway, the remaining steps of this invention (including key parameter screening, prediction model construction, hybrid prediction model construction, and optimization control) can be directly applied. The following detailed description uses high-sulfur wastewater treatment as an example, but this is merely a preferred embodiment to illustrate the technical solution of this invention and is not intended to limit the invention.

[0070] like Figure 1 As shown, the method in this embodiment includes the following steps:

[0071] S1. Establish and calibrate a mechanistic model of microbial sulfur oxidation.

[0072] This step specifically includes:

[0073] S11. Based on the main biological oxidation pathways of sulfides in wastewater treatment (as shown in formulas (3) and (4)), establish a mechanistic model framework (i.e., a biotransformation mechanism model) to describe the microbial kinetics of the conversion process of sulfides, elemental sulfur, and sulfate:

[0074] , formula (3),

[0075] , formula (4).

[0076] Specifically, the mechanistic model framework of the microbial kinetics uses reaction equations that include processes such as sulfide oxidation, elemental sulfur oxidation, and microbial decay, as well as biochemical stoichiometry matrices as shown in Table 1, to characterize the material transformation relationships, and uses kinetic equations in the form of Monod and Haldane to describe the reaction rates.

[0077] Table 1 Biochemical Matrix of Microbial Sulfur Oxidation Mechanism Model

[0078]

[0079] The reaction rate of sulfide oxidation to elemental sulfur As shown in Equation (5), this rate equation comprehensively considers the sulfide substrate inhibition effect and the dissolved oxygen limitation effect, wherein the sulfide inhibition term adopts the Haldane-type expression form, and the dissolved oxygen limitation term adopts the Monod-type expression form:

[0080] , formula (5),

[0081] in, This represents the maximum specific growth rate of sulfur-oxidizing bacteria when sulfides are used as substrates. This refers to the sulfide concentration. Dissolved oxygen concentration, The saturation constant of sulfides is... This is the half-saturation constant of dissolved oxygen during sulfide oxidation. The sulfide inhibition constant is... This refers to the biomass concentration of sulfur-oxidizing bacteria. This is a correction factor used to characterize the proportion of active biomass involved in sulfide oxidation reactions.

[0082] The reaction rate of elemental sulfur oxidation to sulfate As shown in Equation (6), this rate equation introduces a substrate switching term to characterize the non-competitive inhibition of elemental sulfur oxidation by sulfides:

[0083] , formula (6),

[0084] in, This represents the maximum specific growth rate of sulfur-oxidizing bacteria when elemental sulfur is used as a substrate. This refers to the concentration of elemental sulfur. The half-saturation constant of elemental sulfur, This is the half-saturation constant of dissolved oxygen during the oxidation of elemental sulfur. The substrate switching constant is... This is a correction factor used to characterize the proportion of active biomass participating in the elemental sulfur oxidation reaction.

[0085] The rate of decline of sulfur-oxidizing bacteria biomass over time Introducing a first-order kinetic decay model, the reaction rate is shown in equation (7):

[0086] , formula (7),

[0087] Where b is the first-order decay rate constant of sulfur-oxidizing bacteria.

[0088] oxygen supply rate The input flux of dissolved oxygen is described using formula (8) by introducing a gas-liquid mass transfer term:

[0089] , formula (8),

[0090] in, The oxygen volumetric mass transfer coefficient is... This represents the dissolved oxygen saturation value. This represents the instantaneous dissolved oxygen concentration.

[0091] S12. Through laboratory-scale microbial sulfur oxidation sequencing batch experiments, data on the changes in sulfide, elemental sulfur, and sulfate concentrations over time were collected under different influent sulfide concentrations and aeration rates to form multiple sets of time-series calibration datasets.

[0092] Specifically, in this embodiment, four sets of experiments were conducted. Each set of experiments was started after the biofilm had stabilized and continued until the dissolved oxygen concentration in the system returned to saturation. During the experiments, concentration data of sulfides, elemental sulfur, and sulfate were collected at regular intervals to form multiple sets of time-series experimental datasets, which were used to characterize the dynamic changes in the microbial sulfur oxidation process.

[0093] S13. Using the calibration dataset, the kinetic parameters in the mechanism model framework are fitted and calibrated using optimization methods such as genetic algorithms to obtain a calibrated mechanism model that can accurately characterize the dynamic trend of the sulfur oxidation process. The range of values ​​of each kinetic parameter obtained during the calibration process is recorded to form a dataset of calibrated kinetic parameter variation ranges.

[0094] Specifically, the dataset of dynamic parameter variation ranges refers to the set of values ​​obtained after calibrating the mechanism model using experimental data from multiple sets of different operating conditions, reflecting the specific values ​​of each dynamic parameter under different conditions. This dataset contains multiple sets of calibration results for each dynamic parameter, which are used for subsequent statistical analysis of data-driven enhancement, hybrid prediction model construction, and regulation optimization.

[0095] S2. Identify key dynamic parameters.

[0096] This step specifically includes:

[0097] S21. Based on the dynamic parameter variation range dataset obtained in step S13, perform a global sensitivity analysis using the total sensitivity index (Sobol index) for each dynamic parameter to quantify the overall influence of a single parameter and its interaction with other parameters on the model output, thus obtaining the total sensitivity index (S) corresponding to each dynamic parameter. Ti );

[0098] , formula (9),

[0099] in, Indicates the first The total sensitivity index corresponding to each dynamic parameter The output variables of the mechanistic model are represented. This represents the total variance of the model output when all uncertain input parameters vary within their specified range. Indicates a fixed parameter The expected value of the model output variance under the given conditions;

[0100] S22. Based on the specific values ​​(multiple sets of values) of each kinetic parameter obtained under different experimental conditions (multiple sets of experiments in step S12), calculate the coefficient of variation (CV) of each kinetic parameter to objectively assess the degree of self-fluctuation of each kinetic parameter with changes in operating conditions:

[0101] , formula (10),

[0102] in, This represents the average value of the corresponding kinetic parameters under different experimental conditions. The standard deviation of this parameter is used to assess the degree of fluctuation of different dynamic parameters as they change with operating conditions by comparing the magnitude of their coefficients of variation.

[0103] S23. Based on the comprehensive evaluation results of the total sensitivity index and coefficient of variation of each dynamic parameter, key dynamic parameters are selected. This selection is achieved through a comprehensive evaluation of the parameter's sensitivity and volatility. For example, the two indices can be weighted equally and then ranked, with the top-ranked parameters selected as key dynamic parameters.

[0104] Specifically, in this embodiment, firstly, based on the dataset of kinetic parameter variation ranges, a global sensitivity analysis of the Sobol index is conducted on each kinetic parameter in the mechanism model of microbial sulfur oxidation. By sampling each kinetic parameter within its variation range and inputting the sampling results into the mechanism model of microbial sulfur oxidation, the changes in the model output are calculated, thereby obtaining the overall sensitivity (Sobol) index corresponding to each kinetic parameter. This index characterizes the overall influence of each parameter and its interaction with other parameters on the model output. Further, to evaluate the fluctuation characteristics of each kinetic parameter under different experimental operating conditions, a coefficient of variation analysis is performed on the dataset of kinetic parameter variation ranges. By calculating the mean and standard deviation of each kinetic parameter in multiple sets of experimental calibration results, the corresponding coefficient of variation is obtained to characterize the relative fluctuation of each kinetic parameter with changes in operating conditions. Subsequently, by combining the Sobol index analysis results and the coefficient of variation analysis results, an equal-weighted approach is used to comprehensively evaluate the sensitivity and volatility of each kinetic parameter. Three kinetic parameters that significantly affect the output of the microbial sulfur oxidation process and exhibit significant fluctuations with changes in operating conditions are identified and determined as key kinetic parameters for subsequent dynamic enhancement and correction using data-driven methods.

[0105] S3. Obtain the dataset used to build the prediction model.

[0106] This step specifically includes:

[0107] S31. Obtain experimental data related to the transformation of the target pollutant from publicly published literature, and construct an experimental dataset after screening and standardization.

[0108] S32. Input the influent conditions and operating conditions in the experimental dataset into the mechanism model, and output the observation results that match the corresponding experiment through parameter calibration to obtain the corresponding key dynamic parameter dataset.

[0109] Specifically, in this embodiment, experimental data on the microbial sulfur oxidation process are first collected from publicly published literature on microbial sulfur oxidation. This data includes influent sulfide concentration, influent sulfate concentration, experimental apparatus volume, instantaneous oxygen input concentration, hydraulic residence time, influent flow rate, and pH under different operating conditions. The acquired experimental data is screened, units are standardized, and missing or incomplete experimental records are removed. The units of measurement, data formats, and time scales used in different literatures are standardized and converted to construct a microbial sulfur oxidation experimental literature dataset for model analysis. Based on this, the influent conditions, such as influent sulfide concentration and instantaneous oxygen input concentration, from each experiment in the microbial sulfur oxidation experimental literature dataset are used as model inputs. These are substituted into the mechanistic model calibrated in step S1. Through parameter inversion (i.e., adjusting model parameters to match the model output with the corresponding experimental observations in the experimental literature dataset), the specific values ​​of key kinetic parameters corresponding to each experimental condition in the literature are obtained. All inversion results are summarized to form the key kinetic parameter dataset. This process established a sample of mapping relationships between operating conditions and key parameter values, providing a data foundation for subsequent hybrid prediction model construction and parameter enhancement.

[0110] S4. Construct prediction models for key dynamic parameters.

[0111] This step specifically includes:

[0112] S41. Based on data-driven algorithms, construct an initial key dynamic parameter prediction model;

[0113] The data-driven algorithms include eight different types of regression data-driven prediction algorithms, used to characterize the nonlinear variation features of key dynamic parameters under different operating conditions. In this embodiment, through comparative analysis of the prediction performance of different data-driven algorithms, it was found that the data-driven algorithm based on categorical feature enhancement performs best in terms of prediction accuracy and generalization ability. Therefore, the algorithm based on categorical feature enhancement was selected as the final method for constructing the key dynamic parameter prediction model.

[0114] S42. Using the influent conditions and operating conditions of the experimental dataset as input variables and the key dynamic parameters in the corresponding key dynamic parameter dataset as output variables, establish an input dataset and an output dataset, and divide them into a training set and a validation set according to the proportion.

[0115] S43. Use the training set to train the prediction model of step S41, and use the validation set to optimize the model hyperparameters until the prediction error of the model on the validation set reaches the preset condition.

[0116] Specifically, in this embodiment, during model construction, the aforementioned microbial sulfur oxidation experimental literature dataset is used as the input variable data source, and the corresponding key kinetic parameter dataset is used as the output variable data, divided into a training set and a validation set in a 4:1 ratio. The training set is used to train the prediction model, enabling it to learn the mapping relationship between changes in operating conditions and key kinetic parameters. After model training, the validation set is used to evaluate the prediction performance of the model, and the hyperparameter combination is optimized using a grid search method to further improve the model's prediction accuracy and generalization ability. The hyperparameter combination whose prediction error meets the preset accuracy requirements is selected as the optimal configuration of the prediction model.

[0117] S44. Retrain the optimized prediction model using the complete input and output datasets to obtain the final prediction model for the key dynamic parameters.

[0118] S5. Build and run the mechanism-data hybrid prediction model.

[0119] The prediction model of the key kinetic parameters obtained in step S4 is coupled online with the mechanism model of microbial sulfur oxidation obtained in step S1 to construct a hybrid prediction model (mechanism-data hybrid prediction model).

[0120] Specifically, in this embodiment, when the hybrid prediction model is running in real time, the system inputs the influent water quality conditions and operating parameters collected in real time into the prediction model of key kinetic parameters, instantly obtaining dynamic estimates of the key kinetic parameters under the current operating conditions. These dynamic estimates are then input into the mechanism model of microbial sulfur oxidation, replacing the original fixed parameter values ​​of the mechanism model. Using the updated parameter values, combined with the same real-time influent data, the mechanism model performs biochemical reaction simulation calculations at the current moment, outputting predicted values ​​for the concentrations of sulfides, elemental sulfur, etc. This process achieves dynamic and adaptive control of the mechanism model parameters.

[0121] S6. Dynamically predict the effluent water quality (concentration of target pollutants and their transformation products).

[0122] This step specifically includes:

[0123] S61. Real-time acquisition of target pollutant water quality parameters, water quantity parameters, and condition parameters to form online operation data;

[0124] S62. Input the online operation data into the hybrid prediction model to calculate the predicted effluent concentration data of the target pollutant and its transformation products.

[0125] Specifically, in this embodiment, the operating status of the high-sulfur wastewater treatment system is first monitored in real time. Water quality and quantity parameters reflecting influent characteristics and operating conditions are acquired online. These parameters include influent sulfide concentration, influent sulfate concentration, experimental equipment volume, instantaneous oxygen input concentration, hydraulic retention time, influent flow rate, and pH, thus forming online operating data for model calculation. Subsequently, this online operating data is input into the aforementioned constructed hybrid prediction model for microbial sulfur oxidation. The hybrid prediction model performs real-time calculations on the sulfur oxidation process, obtaining predicted effluent concentrations of sulfide, elemental sulfur, and sulfate at the corresponding operating time, used to characterize the dynamic trends of the sulfur oxidation process. The prediction results are as follows: Figure 2 As shown in the figure, this figure illustrates the predicted effluent water quality after 10 rounds of active adjustment of key operating conditions, such as influent sulfide concentration and aeration rate, during a 39-day continuous operation. Figure 2 It is evident that despite multiple changes in operating conditions, the prediction curves of the hybrid prediction model for the concentrations of sulfides, elemental sulfur, and sulfate in the effluent showed good agreement with the corresponding measured values. Particularly under the two significant influent load shocks around days 15 and 25, the model was still able to accurately track the dynamic changes in the concentrations of each component in the effluent, and the prediction error remained at a low level. This fully demonstrates that the mechanism-data hybrid prediction model constructed in this invention can effectively adapt to frequent fluctuations in influent water quality and operating conditions, achieving accurate and stable predictions of the concentrations of key pollutants and their conversion products during sulfur oxidation, providing a reliable input basis for subsequent optimization and control.

[0126] S7, Multi-objective optimization and real-time control.

[0127] This step specifically includes:

[0128] S71. Perform interpretability analysis (such as SHAP analysis) on the input variables in the hybrid prediction model to identify the degree of influence of each input variable (such as aeration rate and internal reflux ratio) on the model prediction results (such as elemental sulfur production), thereby identifying key control parameters (such as aeration rate) that are crucial to process regulation. The identification of key control parameters includes: based on the results of the interpretability analysis, analyzing the contribution of each input variable to the predicted results of sulfide removal rate and elemental sulfur production to identify key control parameters that have a major impact on the model output during sulfur oxidation. These control parameters are used to characterize the regulatory effect of aeration intensity, operating environment conditions, and other adjustable operating variables on the sulfur oxidation process. Specifically, in this embodiment, SHAP value analysis was used, and it was found that the SHAP values ​​of aeration rate and influent pH were much higher than those of other variables; therefore, they were identified as key control parameters.

[0129] S72. Construct a dual-objective optimization model with the optimization objectives of maximizing the removal efficiency of target pollutants and the resource recovery rate of target products.

[0130] In this embodiment, a dual-objective optimization control model for the sulfur oxidation process is constructed, with the maximization of sulfide removal rate and the maximization of elemental sulfur recovery rate as optimization objectives:

[0131] The objective functions for optimization are defined as shown in Equation (11) and Equation (12), respectively:

[0132] , formula (11),

[0133] , formula (12),

[0134] The dual-objective optimization model is defined as shown in formula (13):

[0135] , formula (13),

[0136] in, The percentage of sulfide removal is %. The recovery rate of elemental sulfur is %. The concentration of sulfides in the influent is expressed as mg·S / L. The concentration of sulfides in the effluent is expressed as mg·S / L. The concentration of elemental sulfur produced is (mg·S / L).

[0137] S73. Based on the feasible range of the key control parameters in step S71, the concentration of the target pollutant and its transformation products predicted by the hybrid prediction model, and the preset control target, solve the dual-objective optimization model to obtain the optimal combination of control parameters.

[0138] Specifically, this includes: adjusting the concentration of sulfides in the influent. As known inputs, the key control parameters identified by S71 (such as aeration rate and influent pH) are used as optimization variables during the optimization process, forming the control parameter combination v, and its feasible range serves as the optimization search space. For any given set of parameter values ​​within this space, the hybrid prediction model can dynamically simulate and output the corresponding predicted effluent concentration. and Substituting this dynamic simulation relationship into formulas (11) and (12), the sulfide removal rate can be constructed. and elemental sulfur recovery rate The functional relationship between the control parameter combination v and the control parameter combination v constitutes a bi-objective optimization problem as shown in formula (13).

[0139] The pre-defined control objective is concretized here as constraints or preference settings for the optimization problem. For example, setting a minimum performance constraint that the sulfide removal rate must meet, or assigning different weights to two objective functions to reflect operational preferences. Based on this, a multi-objective optimization algorithm (such as NSGA-II) is used to solve the problem within the feasible space of the control parameters.

[0140] The algorithm outputs a set of Pareto optimal solutions, representing the sulfide removal rate. and elemental sulfur recovery rate The trade-off between these factors (Pareto front). Ultimately, based on preset control objectives and preferences, an optimal combination of control parameters v' (such as the optimal aeration rate and influent pH setpoint) is selected from the Pareto front for real-time process control.

[0141] like Figure 3 As shown, under conditions of 30 days of continuous operation and three changes in influent sulfide concentration gradients, the optimized control strategy based on the output of the hybrid prediction model achieved a synergistic improvement in sulfide removal rate and elemental sulfur recovery rate, reaching an average sulfide removal rate of 96% and an elemental sulfur recovery rate of 91%, respectively. The measured data (i.e., the verification results corresponding to the data points) and the model prediction results (i.e., the calculation results corresponding to the data lines) in the figure show a high degree of agreement, indicating that the optimized control method proposed in this invention can effectively adapt to fluctuations in influent water quality. While ensuring efficient sulfide removal, it greatly improves the recovery efficiency of elemental sulfur, verifying the effectiveness and reliability of the method in a real dynamic operating environment.

[0142] Example 2

[0143] Mechanism-data hybrid driven modeling and control systems for wastewater biochemical treatment processes include:

[0144] The mechanism model building module is used to establish and calibrate the biotransformation mechanism model of the target pollutant, and obtain the calibrated mechanism model and the range of kinetic parameter changes after calibration.

[0145] The key parameter screening module is used to perform sensitivity analysis and coefficient of variation analysis on each dynamic parameter of the mechanism model to screen out key dynamic parameters.

[0146] The prediction model dataset acquisition module is used to acquire experimental datasets of the biotransformation process of the target pollutant and perform parameter inversion based on the mechanism model to obtain the corresponding key kinetic parameter datasets.

[0147] The prediction model building module is used to construct a prediction model for the key dynamic parameters using the experimental dataset as the input dataset and the key dynamic parameter dataset as the output dataset, and by employing a data-driven algorithm.

[0148] A hybrid prediction model construction module is used to couple the prediction model and the mechanism model online to construct a hybrid prediction model. When the hybrid prediction model is running, the prediction model dynamically updates the values ​​of key dynamic parameters in the mechanism model based on real-time water inflow conditions and operating conditions.

[0149] The dynamic prediction module is used to input real-time influent and operating conditions into the mixing prediction model to dynamically predict the concentration of target pollutants and their conversion products.

[0150] The optimization and control module is used to construct and solve a multi-objective optimization problem with the target pollutant removal rate and target product resource recovery rate as optimization objectives based on the prediction results of the hybrid prediction model and the preset control target, to obtain the optimal control parameters and perform real-time control.

[0151] Specifically, in this embodiment, the system is applied to a high-sulfur wastewater treatment scenario. The hybrid prediction model construction module is based on a microbial sulfur oxidation mechanism model and incorporates a data-driven algorithm to enhance the prediction of key kinetic parameters in the microbial sulfur oxidation mechanism model, forming a data-driven enhanced microbial sulfur oxidation hybrid prediction model. This model enables dynamic prediction of the conversion behavior of sulfides, elemental sulfur, and sulfates during microbial sulfur oxidation. The optimization and control module, with sulfide removal efficiency and elemental sulfur recovery efficiency as control objectives, constructs a multi-objective optimization model for the microbial sulfur oxidation process. This model calculates and outputs optimal control parameters based on real-time operating data, preset control objectives, and prediction results. These parameters guide the adjustment of aeration conditions and operating conditions, thereby achieving synergistic optimization of efficient sulfide removal and sulfur resource recovery in high-sulfur wastewater.

[0152] The above embodiments are preferred embodiments of the present invention and are not intended to limit the present invention. Any changes or other equivalent substitutions made without departing from the technical solution of the present invention are included within the protection scope of the present invention.

Claims

1. A mechanism-data hybrid driven modeling and control method for wastewater biochemical treatment processes, characterized in that, Including the following steps: S1. Establish and calibrate a biotransformation mechanism model of the target pollutant to obtain the calibrated mechanism model and the range of kinetic parameter changes after calibration; S2. Perform sensitivity analysis and coefficient of variation analysis on each dynamic parameter of the mechanism model to screen out key dynamic parameters; S3. Obtain the experimental dataset of the biotransformation process of the target pollutant, and perform parameter inversion based on the mechanism model to obtain the corresponding key kinetic parameter dataset; S4. Using the experimental dataset as the input dataset and the key dynamic parameter dataset as the output dataset, a prediction model for the key dynamic parameters is constructed using a data-driven algorithm. S5. Couple the prediction model and the mechanism model online to construct a hybrid prediction model; when the hybrid prediction model is running, based on the real-time water inflow conditions and operating conditions, the prediction model dynamically updates the values ​​of key dynamic parameters in the mechanism model. S6. Input the real-time influent and operating conditions into the mixing prediction model to dynamically predict the concentration of the target pollutants and their conversion products; S7. Based on the prediction results of the hybrid prediction model and the preset control target, construct and solve a multi-objective optimization problem with the target pollutant removal rate and the target product resource recovery rate as optimization objectives, obtain the optimal control parameters, and perform real-time control.

2. The mechanism-data hybrid driven wastewater biochemical treatment process modeling and control method according to claim 1, characterized in that, Step S1 specifically includes: S11. Based on the main biological reaction pathways and transformation relationships of the target pollutants, a mechanistic model framework for microbial dynamics is constructed, which includes reaction equations, biochemical stoichiometry matrices, and microbial dynamics equations. S12. Through laboratory sequential batch experiments, collect concentration data of target pollutants and their transformation products over time under different operating conditions to form a calibration dataset; S13. Using the calibration dataset, fit and calibrate the dynamic parameters in the mechanism model framework to obtain the calibrated mechanism model and the range of variation of the calibrated dynamic parameters.

3. The mechanism-data hybrid driven modeling and control method for wastewater biochemical treatment processes according to claim 1, characterized in that, Step S2 specifically includes: S21. Based on the range of variation of the dynamic parameters, calculate the total sensitivity index of each dynamic parameter to evaluate the degree of influence of each dynamic parameter on the output of the mechanism model; S22. Based on multiple sets of values ​​of the dynamic parameters obtained under different experimental conditions within the range of variation of the dynamic parameters, calculate their coefficient of variation to assess the degree of fluctuation of each dynamic parameter with operating conditions. S23. Based on the comprehensive evaluation results of the total sensitivity index and coefficient of variation of each dynamic parameter, key dynamic parameters are selected.

4. The mechanism-data hybrid driven wastewater biochemical treatment process modeling and control method according to claim 1, characterized in that, Step S3 specifically includes: S31. Obtain experimental data related to the transformation of the target pollutant from publicly published literature, and construct an experimental dataset after screening and standardization. S32. Input the inflow conditions from the experimental dataset into the mechanism model, and output the observation results that match the corresponding experiment through parameter calibration to obtain the corresponding key dynamic parameter dataset.

5. The mechanism-data hybrid driven modeling and control method for wastewater biochemical treatment processes according to claim 1, characterized in that, Step S4 specifically includes: S41. Based on data-driven algorithms, construct an initial key dynamic parameter prediction model; S42. Using the influent conditions and operating conditions of the experimental dataset as input variables and the key dynamic parameters in the corresponding key dynamic parameter dataset as output variables, establish an input dataset and an output dataset, and divide them into a training set and a validation set according to the proportion. S43. Use the training set to train the prediction model of step S41, and use the validation set to optimize the model hyperparameters until the prediction error of the model on the validation set reaches the preset condition. S44. Retrain the optimized prediction model using the complete input and output datasets to obtain the final prediction model for the key dynamic parameters.

6. The mechanism-data hybrid driven modeling and control method for wastewater biochemical treatment processes according to claim 1, characterized in that, In step S4, the data-driven algorithm is a data-driven algorithm based on categorical feature enhancement.

7. The mechanism-data hybrid driven modeling and control method for wastewater biochemical treatment processes according to claim 1, characterized in that, Step S7 specifically includes: S71. Perform interpretability analysis on the input variables in the hybrid prediction model, identify the degree of influence of each input variable on the model prediction results, and identify key control parameters; S72. Construct a dual-objective optimization model with the optimization objectives of maximizing the removal efficiency of target pollutants and the resource recovery rate of target products. S73. Based on the feasible range of the key control parameters in step S71, the concentration of the target pollutant and its transformation products predicted by the hybrid prediction model, and the preset control target, solve the dual-objective optimization model to obtain the optimal combination of control parameters.

8. The mechanism-data hybrid driven modeling and control method for wastewater biochemical treatment processes according to claim 7, characterized in that, The target pollutant removal efficiency and the target product resource recovery rate are calculated according to the following formulas: , formula (1), , Official (2), in, The target pollutant removal efficiency (%). The target product resource recovery rate (%). The target pollutant concentration (mg / L) in the influent. The concentration of the target pollutant in the effluent (mg / L) is given. ν represents the product concentration (mg / L), and v is the control parameter.

9. The mechanism-data hybrid driven modeling and control method for wastewater biochemical treatment processes according to claim 1, characterized in that, The target pollutant is sulfide, the conversion product includes elemental sulfur, and the multi-objective optimization is to simultaneously maximize the sulfide removal rate and the elemental sulfur recovery rate.

10. A mechanism-data hybrid driven modeling and control system for wastewater biochemical treatment processes, characterized in that, include: The mechanism model building module is used to establish and calibrate the biotransformation mechanism model of the target pollutant, and obtain the calibrated mechanism model and the range of kinetic parameter changes after calibration. The key parameter screening module is used to perform sensitivity analysis and coefficient of variation analysis on each dynamic parameter of the mechanism model to screen out key dynamic parameters. The prediction model dataset acquisition module is used to acquire experimental datasets of the biotransformation process of the target pollutant and perform parameter inversion based on the mechanism model to obtain the corresponding key kinetic parameter datasets. The prediction model building module is used to construct a prediction model for the key dynamic parameters using the experimental dataset as the input dataset and the key dynamic parameter dataset as the output dataset, and by employing a data-driven algorithm. A hybrid prediction model construction module is used to couple the prediction model and the mechanism model online to construct a hybrid prediction model. When the hybrid prediction model is running, the prediction model dynamically updates the values ​​of key dynamic parameters in the mechanism model based on real-time water inflow conditions and operating conditions. The dynamic prediction module is used to input real-time influent and operating conditions into the mixing prediction model to dynamically predict the concentration of target pollutants and their conversion products. The optimization and control module is used to construct and solve a multi-objective optimization problem with the target pollutant removal rate and target product resource recovery rate as optimization objectives based on the prediction results of the hybrid prediction model and the preset control target, to obtain the optimal control parameters and perform real-time control.

Images