A Coupled Control System and Method for Deep Denitrification of Wastewater

Abstract

This invention relates to the field of wastewater treatment technology, and in particular to a coupled control system and method for deep denitrification of wastewater, comprising a real-time water quality monitoring module, a microbial twin modeling module, a real-time prediction module, an intelligent decision-making module, an actuator module, and a predictive control module. Compared to existing technologies that typically employ passive feedback control strategies based on effluent water quality indicators, which suffer from strong lag and are unable to cope with drastic fluctuations in influent load, this invention elevates the control objective from traditional process parameter setpoints to direct optimization of microbial ecological functions by constructing a digital twin capable of reflecting the functional state of the microbial community in real time. This represents a fundamental leap from "controlling process parameters" to "regulating microbial ecology," enabling intervention at the root of the reaction process and giving the system proactive health management capabilities, thereby significantly improving the stability of treatment efficiency and the level of intelligence in handling complex operating conditions.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, and in particular to a coupled control system and method for deep denitrification of wastewater. Background Technology

[0002] Biological nitrogen removal from wastewater is a core process in modern wastewater treatment plants, and its efficiency directly affects the ecological and environmental safety of receiving water bodies. With increasingly stringent emission standards and ever-improving energy conservation and emission reduction requirements, achieving deep, stable, and low-energy removal of nitrogen pollutants has become an urgent industry need. Currently, this process mainly relies on the metabolic activity of specific microbial communities (such as nitrifying and denitrifying bacteria) in activated sludge. Their activity is easily affected by complex factors such as influent load, temperature, and pH, thus placing extremely high demands on process control.

[0003] Most existing control technologies remain at the feedback or feedforward control level based on key water quality parameters (such as ammonia nitrogen, nitrate nitrogen, and dissolved oxygen). For example, they control the dissolved oxygen concentration in the aerobic zone by adjusting the aeration rate, or add carbon sources based on the influent flow rate and nitrate nitrogen concentration. These methods have significant limitations: First, their control behavior is lagging, belonging to "post-event remediation," and is difficult to effectively cope with the impact of drastic fluctuations in influent water quality; second, the control models usually rely on simplified empirical formulas or fixed kinetic parameters, which cannot accurately describe the dynamic changes and activity status of the microbial community in real wastewater treatment systems, resulting in insufficient model prediction accuracy and limited optimization effects; finally, existing technologies generally limit the control dimension to a few operational variables, while neglecting the active optimization and regulation of environmental parameters such as temperature and pH that have a key impact on microbial activity, thus limiting the further exploration of denitrification potential.

[0004] To address the aforementioned problems, this invention proposes a coupled control system and method for deep nitrogen removal from wastewater. The core idea of ​​this invention is to construct a "microbial digital twin" synchronized in real time with the physical-biological treatment tank. By integrating mechanistic knowledge with real-time data, it dynamically tracks and predicts the intangible microbial activity state. Based on this, the system not only optimizes traditional aeration and carbon source addition strategies but also incorporates environmental parameters such as temperature and pH into the synergistic optimization framework. Furthermore, it utilizes influent load prediction for proactive regulation, ultimately forming a closed-loop intelligent control system that precisely intervenes at the microbial ecological level. This aims to achieve stable effluent quality compliance, optimized energy and material consumption, and a comprehensive improvement in the system's resilience to shock loads. Summary of the Invention

[0005] To overcome the problems mentioned in the background art, the present invention proposes a coupled control system and method for deep denitrification of wastewater.

[0006] The technical solution of this invention is: a coupled control system for deep denitrification of wastewater, comprising:

[0007] The real-time water quality monitoring module is used to collect water quality parameters in the biological treatment tank in real time.

[0008] The microbial twin modeling module is used to build and dynamically update digital twin models that reflect microbial activity based on water quality parameters.

[0009] The real-time prediction module is used to obtain the prediction results of the microbial state based on the digital twin model, current water inflow data and future water inflow prediction;

[0010] The intelligent decision-making module is used to generate optimal control commands based on the predicted results of microbial status and digital twin models;

[0011] The actuator module, connected to the intelligent decision module, is used to execute control commands and adjust the aeration equipment and carbon source dosing equipment.

[0012] The prediction and control module is used to analyze historical influent data to predict the future changes in influent water quality and quantity, and provides the prediction results to the intelligent decision-making module for advance control.

[0013] As a preferred option, the microbial twin modeling module specifically includes:

[0014] The mechanism model library stores mathematical models that describe the biochemical reaction process of wastewater. These mathematical models contain key parameters that represent the physiological state of the microbial community.

[0015] The hybrid model engine is used to build and run mechanism-data driven hybrid models, which consist of parallel mechanistic model sub-units and data-driven residual model sub-units.

[0016] The data assimilation engine is used to receive real-time water quality data and use data assimilation algorithms to fuse and dynamically correct the real-time water quality data with the mechanism model.

[0017] Microbial state estimator is used to output real-time estimates of microbial activity indicators that cannot be directly measured, based on a calibrated model, thereby constructing and updating a digital twin that reflects the true state of microorganisms.

[0018] As a preferred embodiment, the prediction process of the hybrid model engine is defined by the following formula:

[0019] ;

[0020] in, The predicted system state vector at time k includes measurable water quality concentration variables and indirect microbial physiological parameters. As a sub-unit of the mechanistic model, it describes biochemical reaction processes based on first-principles calculations. Let k be the state vector at time k-1. For real-time system inputs, including influent flow rate and influent water quality, This is the set of inherent parameters for the mechanistic model. This is a data-driven residual model subunit used to learn the predicted residuals of the mechanism model subunit. This refers to the system's historical operating data up to time k-1. These are the adjustable parameters for the residual model sub-units.

[0021] As a preferred choice, the data assimilation engine uses the ensemble Kalman filter algorithm, whose execution steps are as follows:

[0022] Prediction step: Based on the state vector of the previous moment, forward simulation is performed through the mechanism model to obtain the predicted value of the state vector at the current moment and its uncertainty;

[0023] Update step: Use the real-time water quality monitoring data at the current moment as the observed value, compare it with the model prediction value, calculate the Kalman gain, and make the optimal correction to the state vector prediction value to obtain the most likely state vector estimate at the current moment;

[0024] Among them, the adjustable parameters of the data-driven residual model subunit in the state vector of the ensemble Kalman filter algorithm are... As part of the state vector, it is estimated and updated together with the microbial physiological parameters, thereby enabling the online self-learning of the residual model subunit.

[0025] As a preferred embodiment, the real-time prediction module specifically includes:

[0026] The prediction initialization unit is used to obtain the real-time corrected digital twin model and its current state vector from the microbial twin modeling module as the initial state for prediction.

[0027] A forward simulation engine is used to drive the digital twin model to perform dynamic simulations along a future timeline;

[0028] The prediction result generator is used to output predicted data on the microbial state and effluent water quality over a future period based on simulation results.

[0029] Preferably, the simulation process of the forward simulation engine is defined by the following formula:

[0030] ;

[0031] in, The model state vector at time k is provided by the microbial twin modeling module and includes microbial physiological parameters and water quality concentration. Forward simulation function of digital twin model, The input sequence is the sequence of future H time steps starting from the current time k, containing future influent water quality and quantity data provided by the prediction and control module. For the set of model parameters, The input sequence is a sequence of state vectors for the next H time steps, predicted based on information at time k. In addition to future water inflow data, the input sequence also includes the control command sequence generated by the intelligent decision-making module as a known input to predict the system response under different control strategies.

[0032] As a preferred embodiment, the intelligent decision-making module, when in operation, specifically includes:

[0033] S11: Obtain the current state vector of the digital twin model that has been corrected in real time by the microbial twin modeling module, as well as the system state prediction sequence in the future prediction time domain provided by the real-time prediction module;

[0034] S12: Based on the current state and the state prediction sequence, construct a finite-time optimization problem with the future control command sequence as the decision variable;

[0035] S13: Solve the optimization problem to obtain the optimal control command sequence;

[0036] S14: Apply the first instruction in the optimal control instruction sequence to the actuator;

[0037] S15: In the next control cycle, based on the new system state measurement value, repeat steps S11 to S14.

[0038] Preferably, the intelligent decision-making module is also used to generate control instructions for microbial environmental parameters, including temperature and pH, based on the digital twin model, in order to optimize the microbial state, specifically:

[0039] S21: Based on the current microbial physiological parameters and target effluent quality obtained from the microbial twin modeling module, determine the optimal activity range of key functional microbial communities and the corresponding target values ​​of optimal environmental parameters.

[0040] S22: Using a digital twin model, the quantitative impact of minute changes in environmental parameters on microbial activity and effluent quality is analyzed through simulation, and a sensitivity matrix for environmental parameter regulation is established.

[0041] S23: Environmental parameters are used as decision variables and incorporated together with operating parameters into the model predictive control framework for collaborative optimization and solution, generating a comprehensive control command sequence;

[0042] S24: Extract environmental parameter control instructions from the optimal control instruction sequence, verify whether they are within the safe operating range of the equipment, and finally send the verified instructions to the actuator module.

[0043] Preferably, the predictive control module, when in operation, specifically includes:

[0044] S31: Analyze long-term historical inflow data to identify the periodic variation patterns of inflow flow and water quality parameters, including daily, weekly, and seasonal cycles, and extract key features;

[0045] S32: Based on the extracted features, establish influent water quality and quantity prediction models for short-term, medium-term and long-term water use respectively;

[0046] S33: Match real-time water inflow data with historical patterns to drive the prediction model to generate a predicted trajectory of future water inflow load and quantify the range of uncertainty in the prediction.

[0047] S34: Provide the inflow prediction results and their uncertainty range to the intelligent decision-making module to generate advance control instructions for future load changes.

[0048] A coupled control method for deep denitrification of wastewater includes the following steps:

[0049] S41: Key parameters in the biological treatment tank are obtained through real-time water quality monitoring, and standardized data are obtained after preprocessing.

[0050] S42: Based on standardized data, dynamically estimate microbial activity using fusion mechanisms and data-driven digital twin models;

[0051] S43: Combining the current system status with future influent predictions, generate prediction results of microbial status and effluent water quality through forward simulation;

[0052] S44: With multiple objectives of achieving effluent quality standards, minimizing energy consumption, and maximizing microbial activity, a rolling optimization algorithm is used to solve for the comprehensive control commands of aeration, carbon source addition, and environmental parameters.

[0053] S45: Execute control commands and perform closed-loop adjustment of equipment operating status;

[0054] S46: At the same time, based on the periodic patterns of historical water inflow data, future load fluctuations can be predicted, enabling advance adjustments to the control strategy.

[0055] The beneficial effects of this invention are:

[0056] 1. Compared to existing technologies that typically employ passive feedback control strategies based on effluent quality indicators, this method suffers from strong lag and cannot cope with drastic fluctuations in influent load, essentially acting as a "post-event remedy." This invention constructs a digital twin that can reflect the functional state of the microbial community in real time, elevating the control objective from traditional process parameter setpoints to direct optimization of microbial ecological functions. This represents a fundamental leap from "controlling process parameters" to "regulating microbial ecology," enabling intervention at the root of the reaction process and giving the system proactive health management capabilities. Consequently, it significantly improves the stability of treatment efficiency and the level of intelligence in handling complex operating conditions.

[0057] 2. Compared to existing technologies where mathematical models often rely on fixed mechanistic models or purely black-box data-driven models, the former is prone to distortion under complex water conditions, while the latter has poor interpretability and weak generalization ability. This invention adopts a hybrid modeling architecture that combines mechanistic and data-driven models in parallel. The mechanistic unit describes known biochemical reaction kinetics to maintain physical meaning, while the data-driven unit learns online and compensates for residuals that the mechanistic model cannot accurately describe. This approach combines the reliability of mechanistic models with the adaptability of data-driven models, and through advanced data assimilation technology, the model parameters can be continuously updated. This ensures that the digital twin can track the dynamic changes of real microbial systems with high fidelity, providing an unprecedentedly reliable model foundation for precise control.

[0058] 3. Compared to existing technologies that typically only use aeration rate and carbon source dosage as operational variables, neglecting the decisive influence of key environmental parameters such as temperature and pH on microbial activity, resulting in limited depth of regulation, this invention incorporates temperature and pH as key decision variables into a model predictive control framework for collaborative optimization. It can dynamically create the optimal reaction environment for functional microbial communities based on influent load and seasonal changes in water temperature, thereby deeply stimulating their metabolic activity. This not only breaks through the traditional approach of only coarsely stabilizing environmental parameters, achieving an upgrade from "coarse maintenance" to "precise optimization," but also fundamentally improves the efficiency and stability of the denitrification reaction, especially under adverse conditions such as low temperature or abnormal influent water quality.

[0059] 4. Compared to existing technologies that focus primarily on improving local aspects or simply connecting modules, lacking the ability for collaborative optimization across the entire process and making it difficult to form a closed-loop decision-making system, this invention constructs a complete closed-loop coupled system from precise sensing, twin modeling, real-time prediction, intelligent decision-making to forward-looking regulation. It organically integrates various highly intelligent modules, enabling the system to not only perform real-time optimization based on the current state but also make forward-looking decisions based on influent prediction, and continuously evolve through rolling optimization and online self-learning mechanisms. This highly coupled, full-chain design breaks down information silos, achieving deep interaction between sensing, decision-making, and execution, ultimately making the entire system an organic whole capable of autonomous adaptation, proactive response, and continuous optimization, comprehensively improving the reliability of effluent compliance, operational economy, and shock resistance. Attached Figure Description

[0060] Figure 1 The diagram shown is a schematic representation of the coupled control system for deep denitrification of wastewater according to the present invention.

[0061] Figure 2 The diagram shown is a schematic flow chart of the coupled control method for deep denitrification of wastewater according to the present invention. Detailed Implementation

[0062] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0063] Please see Figure 1 The present invention provides an embodiment: a coupled control system for deep denitrification of wastewater, comprising:

[0064] I. Real-time Water Quality Monitoring Module

[0065] Used to collect water quality parameters in the biochemical pool in real time.

[0066] Specifically, the real-time water quality monitoring module includes:

[0067] A sensor array, deployed within the biochemical reaction tank, is used for continuous, online measurement of various key water quality parameters related to the biological denitrification process;

[0068] The data acquisition and preprocessing unit is connected to the sensor array signal and is used to receive raw sensor signals and perform preprocessing operations such as filtering, noise reduction and outlier removal to generate standardized and reliable water quality data.

[0069] The data communication unit, connected to the data acquisition and preprocessing unit, is used to transmit the preprocessed water quality data to the microbial twin modeling module in real time.

[0070] Preferably, the sensor array includes: an ammonia nitrogen analyzer for measuring ammonia nitrogen concentration; a nitrate nitrogen analyzer for measuring nitrate nitrogen and nitrite nitrogen concentration; a dissolved oxygen sensor for measuring dissolved oxygen concentration; a pH sensor for measuring the acidity or alkalinity of the mixture; a temperature sensor for measuring the temperature of the mixture; a redox potential sensor for indirectly reflecting the anoxic and anaerobic states of the biochemical environment; and a chemical oxygen demand or total organic carbon analyzer for measuring the organic matter content in the water.

[0071] Preferably, the sensor array is deployed at multiple points, with its measuring points covering the anaerobic, anoxic, and aerobic zones of the biochemical reactor to obtain process parameters that change along the process.

[0072] Preferably, the data acquisition and preprocessing unit is further configured to perform preliminary calculations on the preprocessed water quality data to obtain the rate of change of water quality parameters, and transmit the rate of change data to the microbial twin modeling module through the data communication unit so that it can establish a quantitative relationship between the rate of change of water quality parameters and microbial activity.

[0073] Preferably, the preprocessing operation performed by the data acquisition and preprocessing unit also includes data validity verification. The verification logic is as follows: compare the current data acquisition with the historical data change trend and the reasonable range based on the process mechanism. If the data exceeds the preset threshold, the data is marked as invalid or suspicious.

[0074] In this embodiment, the present invention deploys a multi-point sensor array covering the anaerobic, anoxic, and aerobic functional zones of the biological treatment tank, continuously measuring key parameters such as ammonia nitrogen, nitrate nitrogen, dissolved oxygen, pH, and temperature online. The raw signals are filtered, denoised, outlier removed, and their validity verified based on the process mechanism by a data acquisition and preprocessing unit, and the rate of change of water quality parameters is calculated. Finally, standardized and reliable data and the rate of change are transmitted to the back-end model in real time. Thus, through high-frequency, multi-point precise sensing and strict data quality control, a highly reliable data foundation is provided for subsequent modeling; in particular, the calculation and provision of the rate of change of water quality parameters directly supports real-time soft measurement of microbial activity, laying a solid foundation for the precise sensing and optimized control of the entire system.

[0075] II. Microbial Twin Modeling Module

[0076] Used to establish and dynamically update digital twin models reflecting microbial activity based on water quality parameters;

[0077] In this embodiment, the microbial twin modeling module specifically includes:

[0078] The mechanism model library stores mathematical models that describe the biochemical reaction process of wastewater. These mathematical models contain key parameters that represent the physiological state of the microbial community.

[0079] The hybrid model engine is used to build and run mechanism-data driven hybrid models, which consist of parallel mechanistic model sub-units and data-driven residual model sub-units.

[0080] The data assimilation engine is used to receive real-time water quality data and use data assimilation algorithms to fuse and dynamically correct the real-time water quality data with the mechanism model.

[0081] Microbial state estimator is used to output real-time estimates of microbial activity indicators that cannot be directly measured, based on a calibrated model, thereby constructing and updating a digital twin that reflects the true state of microorganisms.

[0082] Specifically, this module provides the theoretical foundation for biochemical reactions through a mechanistic model library. It utilizes a hybrid model engine to construct a parallel mechanistic-data-driven hybrid model (where the mechanistic sub-unit describes known kinetic laws, and the data-driven sub-unit learns model residuals). Then, a data assimilation engine dynamically fuses and corrects real-time water quality monitoring data with the hybrid model. Finally, a microbial state estimator enables real-time soft measurement of microbial activity indicators that cannot be directly measured. By leveraging the complementary advantages of mechanistic and data-driven models, the module maintains model interpretability while improving prediction accuracy under complex real-world conditions. Data assimilation technology enables real-time dynamic tracking of microbial activity parameters, allowing the digital twin to continuously and synchronously reflect the true state of the microbial system. This provides a reliable model foundation for subsequent precise control and optimized regulation, ultimately significantly improving the operational efficiency and shock load resistance of the wastewater denitrification system.

[0083] In this embodiment, the prediction process of the hybrid model engine is defined by the following formula:

[0084] ;

[0085] in, The predicted system state vector at time k includes measurable water quality concentration variables and indirect microbial physiological parameters. As a sub-unit of the mechanistic model, it describes biochemical reaction processes based on first-principles calculations. Let k be the state vector at time k-1. For real-time system inputs, including influent flow rate and influent water quality, This is the set of inherent parameters for the mechanistic model. This is a data-driven residual model subunit used to learn the predicted residuals of the mechanism model subunit. This refers to the system's historical operating data up to time k-1. These are the adjustable parameters for the residual model sub-units.

[0086] Specifically, data-driven residual model subunit Implemented using one of the following algorithms: neural network, Gaussian process regression, or support vector machine, with its input dimension expanded to include a sequence of historical states. To capture the dynamic lag effect and unmodeled periodicity of the system.

[0087] In this embodiment, the data assimilation engine uses the ensemble Kalman filter algorithm, and its execution steps are as follows:

[0088] Prediction step: Based on the state vector of the previous moment, forward simulation is performed through the mechanism model to obtain the predicted value of the state vector at the current moment and its uncertainty;

[0089] Update step: Use the real-time water quality monitoring data at the current moment as the observed value, compare it with the model prediction value, calculate the Kalman gain, and make the optimal correction to the state vector prediction value to obtain the most likely state vector estimate at the current moment;

[0090] Among them, the adjustable parameters of the data-driven residual model subunit in the state vector of the ensemble Kalman filter algorithm are... As part of the state vector, it is estimated and updated together with the microbial physiological parameters, thereby enabling the online self-learning of the residual model subunit.

[0091] Specifically, in the microbial digital twin modeling, this module employs an ensemble Kalman filter algorithm as the core of the data assimilation engine. It performs forward state simulation using a mechanistic model in a "prediction step," and then uses real-time monitoring data as observations to optimally correct the prediction results in an "update step." The algorithm's unique feature lies in incorporating the adjustable parameter Φ of the data-driven residual model along with microbial physiological parameters into the state vector for joint estimation and real-time updating. This achieves continuous synchronous calibration between the digital twin model and the real system state, significantly improving the model's dynamic tracking accuracy under complex real-world conditions. Simultaneously, through online self-learning of parameter Φ, the hybrid model possesses the ability to autonomously evolve to adapt to long-term system changes, providing a highly adaptive and reliable model foundation for subsequent accurate prediction and optimized control.

[0092] Preferably, the microbial physiological parameters include the maximum specific growth rate of nitrifying bacteria and the maximum specific growth rate of denitrifying bacteria, and the mechanistic model subunit. The following kinetic equation correlates microbial activity with water quality changes:

[0093] ;

[0094] in, This refers to the substrate consumption or product formation rate. For the corresponding microbial physiological parameters, Microbial concentration, This is the microbial yield coefficient. Substrate concentration, It is the half-saturation constant. This is an environmental inhibitory factor item;

[0095] Furthermore, by assimilating real-time measured water quality change rate data, the state vector is... and data-driven residual model subunit The outputs are jointly estimated.

[0096] Specifically, the models stored in the mechanism model library are activated sludge model series or their simplified and modified forms.

[0097] This embodiment provides a microbial twin modeling method for deep denitrification of wastewater. Its core lies in constructing a hybrid model that integrates mechanistic and data-driven approaches. The mechanistic model sub-unit describes biochemical reaction kinetics based on Monod's equations, while the data-driven sub-unit learns residuals that the mechanistic model cannot capture. An ensemble Kalman filter algorithm is used as the data assimilation engine to dynamically fuse and correct real-time water quality monitoring data with the hybrid model. Simultaneously, adjustable parameters of the data-driven sub-unit, along with key microbial physiological parameters, are used as state vectors for real-time estimation and online self-learning. This achieves complementary advantages through mechanistic and data-driven approaches, significantly improving the model's prediction accuracy and generalization ability in complex real-world environments. Data assimilation technology enables "real-time soft measurement" of microbial activities that cannot be directly measured (such as nitrifying and denitrifying bacteria activity). Finally, a digital twin capable of dynamically and synchronously reflecting the true state of microorganisms with high fidelity is constructed, providing a reliable model foundation for subsequent accurate prediction and optimized control, achieving a leap from "controlling process parameters" to "regulating microbial ecology."

[0098] III. Real-time Prediction Module

[0099] Used to obtain prediction results of microbial status based on digital twin models, current influent data and future influent prediction;

[0100] In this embodiment, the real-time prediction module specifically includes:

[0101] The prediction initialization unit is used to obtain the real-time corrected digital twin model and its current state vector from the microbial twin modeling module as the initial state for prediction.

[0102] A forward simulation engine is used to drive the digital twin model to perform dynamic simulations along a future timeline;

[0103] The prediction result generator is used to output predicted data on the microbial state and effluent water quality over a future period based on simulation results.

[0104] In this embodiment, the simulation process of the forward simulation engine is defined by the following formula:

[0105] ;

[0106] in, The model state vector at time k is provided by the microbial twin modeling module and includes microbial physiological parameters and water quality concentration. Forward simulation function of digital twin model, The input sequence is the sequence of future H time steps starting from the current time k, containing future influent water quality and quantity data provided by the prediction and control module. For the set of model parameters, The input sequence is a sequence of state vectors for the next H time steps, predicted based on information at time k. In addition to future water inflow data, the input sequence also includes the control command sequence generated by the intelligent decision-making module as a known input to predict the system response under different control strategies.

[0107] Preferably, the forward simulation engine uses a rolling time-domain method for simulation, that is, at each sampling time, it receives the latest initial state and the updated future input sequence, and performs a completely new future H-step prediction to achieve online rolling updates of the prediction results.

[0108] Preferably, the forward simulation engine performs multi-scenario prediction, which constructs multiple initial states and input sequences representing different uncertainties, and runs multiple simulation instances in parallel to obtain the prediction range of future states rather than just a single point prediction value.

[0109] Specifically, the prediction data output by the prediction result generator includes:

[0110] Microbial activity prediction trajectory: key microbial physiological parameters in the near future;

[0111] Effluent water quality prediction trajectory: Concentration changes of key effluent indicators and early warning of whether they exceed standards in the future;

[0112] Multi-scenario prediction boundary: The probability distribution and confidence interval of the above trajectory obtained from multi-scenario prediction.

[0113] In this embodiment, the present invention utilizes a digital twin model that has been calibrated in real time. Starting with the current system state as the initial condition and incorporating future influent predictions and control command sequences, it dynamically extrapolates through a forward simulation function. This process employs a rolling time-domain method to continuously update predictions and generates probabilistic prediction results based on parallel simulations across multiple scenarios, including microbial activity trajectories, effluent water quality changes, and confidence intervals. This transforms traditional passive response into proactive look-ahead control. By using high-fidelity simulation to predict the system's behavior under various future operating conditions, it not only provides crucial information for optimization decisions but also significantly enhances the system's robustness against influent fluctuations by quantifying prediction uncertainties. This achieves proactive protection of effluent water quality and active prevention of operational risks.

[0114] IV. Intelligent Decision-Making Module

[0115] Used to generate optimal control commands based on microbial state-based predictions and digital twin models.

[0116] In this embodiment, the intelligent decision-making module, when in operation, specifically includes:

[0117] S11: Obtain the current state vector of the digital twin model that has been corrected in real time by the microbial twin modeling module, as well as the system state prediction sequence in the future prediction time domain provided by the real-time prediction module;

[0118] S12: Based on the current state and the state prediction sequence, construct a finite-time optimization problem with the future control command sequence as the decision variable;

[0119] S13: Solve the optimization problem to obtain the optimal control command sequence;

[0120] S14: Apply the first instruction in the optimal control instruction sequence to the actuator;

[0121] S15: In the next control cycle, based on the new system state measurement value, repeat steps S11 to S14.

[0122] In this embodiment, when constructing a finite-time domain optimization problem with future control command sequences as decision variables based on the current state and the state prediction sequence, the constructed optimization problem is defined by minimizing the following objective function:

[0123] ;

[0124] in, To output the tracking item, To control penalty items, For microbial activity optimization, For the current moment, For the prediction step index, To predict the time domain, Let k be the system output variable predicted at time k+i. These are reference values ​​for the output variables, i.e., the desired effluent water quality indicators. To output the weight matrix of the tracking term, To control the time domain, This is the control increment at time k+i calculated at time k. To control the weight matrix of the incremental penalty term, These are the key physiological parameters of the microorganisms predicted at time k+i, representing the activity of the microorganisms. This represents the optimal value for the microbial physiological parameters. This is the weight matrix for the microbial activity optimization term.

[0125] Specifically, the optimal value in the microbial activity optimization term. It is not a fixed setpoint, but a time-varying target range calculated by the system based on influent load, water temperature, and sludge age. Its purpose is to maintain the activity of the microbial community within a range that ensures treatment efficiency while being energy-saving and economical.

[0126] Specifically, the weight matrix , and It can adaptively adjust according to the changing characteristics of the influent load; during periods of high load impact, the weight of the output tracking term is increased. Prioritize ensuring effluent meets standards; during stable operation periods, increase the weight of the microbial activity optimization item. Weights of control penalty items Prioritize energy conservation, consumption reduction, and stable operation.

[0127] Specifically, the optimization problem-solving process incorporates microbial environmental parameters, including temperature and pH, as part of the optimization decision variable U, and obtains the environmental parameter control instructions required to create optimal reaction conditions for the microorganisms.

[0128] In this embodiment, the intelligent decision-making module is also used to generate control instructions for microbial environmental parameters, including temperature and pH value, based on the digital twin model, so as to optimize the microbial state, specifically:

[0129] S21: Based on the current microbial physiological parameters and target effluent quality obtained from the microbial twin modeling module, determine the optimal activity range of key functional microbial communities and the corresponding target values ​​of optimal environmental parameters.

[0130] S22: Using a digital twin model, the quantitative impact of minute changes in environmental parameters on microbial activity and effluent quality is analyzed through simulation, and a sensitivity matrix for environmental parameter regulation is established.

[0131] S23: Environmental parameters are used as decision variables and incorporated together with operating parameters into the model predictive control framework for collaborative optimization and solution, generating a comprehensive control command sequence;

[0132] S24: Extract environmental parameter control instructions from the optimal control instruction sequence, verify whether they are within the safe operating range of the equipment, and finally send the verified instructions to the actuator module.

[0133] Specifically, when determining the optimal activity range and corresponding target values ​​of key functional microbial communities based on the current microbial physiological parameters and target effluent quality obtained from the microbial twin modeling module, the target values ​​of the optimal environmental parameters are not fixed constants, but are dynamically determined by querying a preset microbial activity-environmental parameter response surface. This response surface is quantitatively described by the following formula:

[0134] ;

[0135] in, For the optimal growth rate, As a fundamental function of microbial activity, For temperature, pH level Substrate concentration, Microbial concentration, Other inhibitory factors.

[0136] When environmental parameters are included as decision variables and incorporated together with operational parameters into the model predictive control framework for collaborative optimization to generate a comprehensive control command sequence, the objective function of the optimization problem adds an environmental parameter adjustment cost term to the standard model predictive control objective function. The formula is as follows:

[0137] ;

[0138] in, To optimize the objective function of the problem, The standard objective function includes effluent water quality, microbial activity, and control rate of change. These are the standard settings for environmental parameters. To control the time domain, Let k be the predicted value of the environmental parameter control command for a future time k+i at time k. The weight matrix for controlling environmental parameter deviations from normal values.

[0139] In the preferred embodiment of the above scheme, the optimization solution process must satisfy the following environmental parameter-related constraints:

[0140] Dynamic constraints: (System dynamics are affected by environmental parameters);

[0141] Path constraints: (Environmental parameters are within their adjustable range);

[0142] Rate constraints: (The rate of change of environmental parameters does not exceed the equipment limits);

[0143] in, This is the predicted value of the system state vector at the next time step k+1. Let k be the system state vector at the current time k. Let k be the operation control variable at time k. These are the environmental parameter control variables at time k, including temperature control commands and pH control commands. For the system dynamics model function, This represents the lower limit of the temperature parameter. This represents the upper limit of the temperature parameter. Let k be the temperature value predicted at time k+i in the future. This is the minimum allowable pH value. This is the maximum allowable pH value. Let pH be the predicted future time k+i at time k. This represents the change in temperature command between two adjacent control cycles. This represents the maximum allowable temperature change within a single control cycle. This represents the change in pH command between two adjacent control cycles. This represents the maximum allowable change in pH value within a single control cycle.

[0144] Preferably, when the intelligent decision-making module generates regulatory instructions for microbial environmental parameters, including temperature and pH, based on the digital twin model to optimize the microbial state, it also includes an assessment of the effectiveness of environmental parameter regulation.

[0145] S25: After the command is executed, the actual changes in microbial activity are continuously monitored, and the actual changes are compared with the predicted changes. Based on the comparison results, the parameters of the sensitivity matrix or the microbial activity-environmental parameter response surface are adaptively adjusted to realize online self-learning of the environmental parameter regulation strategy.

[0146] In this embodiment, the present invention employs a rolling optimization framework based on model predictive control (MPC). It solves for the optimal control commands for aeration rate, carbon source dosage, and environmental parameters such as temperature and pH by minimizing a multi-objective function that includes effluent quality tracking, control penalty, and microbial activity optimization terms. This method innovatively incorporates environmental parameters as decision variables into the optimization solution and dynamically determines the optimal activity target based on the microbial activity-environmental parameter response surface. Furthermore, it introduces an environmental parameter adjustment cost term into the objective function to balance the control effect with economic efficiency. All optimization solutions satisfy kinetic, path, and rate constraints to ensure safety and feasibility. This approach achieves a leap from simply ensuring effluent meets standards to proactively optimizing microbial ecological functions. By taking microbial activity as a direct optimization target, it fundamentally improves the system's treatment efficiency and stability. Secondly, through multi-objective rolling optimization and adaptive weight adjustment, it prioritizes ensuring effluent meets standards during high-load shocks and automatically optimizes energy consumption during stable periods, achieving a dynamic balance between treatment effectiveness and operational economy. Finally, by incorporating environmental parameters into the collaborative optimization framework, it breaks through the limitations of traditional control methods that only adjust aeration and carbon sources. By creating the optimal microbial reaction environment, it significantly improves the depth of denitrification efficiency regulation and the system's robustness in handling complex operating conditions.

[0147] V. Execution Mechanism Module

[0148] It connects to the intelligent decision-making module to execute control commands and adjust the aeration equipment and carbon source dosing equipment;

[0149] Specifically, the actuator module includes one or more of the following: a variable frequency blower, an air flow regulating valve, a carbon source dosing pump, an acid-base dosing pump, and a heat exchanger.

[0150] VI. Prediction and Control Module

[0151] It is used to analyze historical influent data to predict future changes in influent water quality and quantity, and to provide the prediction results to the intelligent decision-making module for advance control.

[0152] In this embodiment, the predictive control module, when in operation, specifically includes:

[0153] S31: Analyze long-term historical inflow data to identify the periodic variation patterns of inflow flow and water quality parameters, including daily, weekly, and seasonal cycles, and extract key features;

[0154] S32: Based on the extracted features, establish influent water quality and quantity prediction models for short-term, medium-term and long-term water use respectively;

[0155] S33: Match real-time water inflow data with historical patterns to drive the prediction model to generate a predicted trajectory of future water inflow load and quantify the range of uncertainty in the prediction.

[0156] S34: Provide the inflow prediction results and their uncertainty range to the intelligent decision-making module to generate advance control instructions for future load changes.

[0157] In this embodiment, when analyzing long-term historical influent data and identifying the periodic variation patterns of influent flow rate and water quality parameters, the periodic variation patterns are identified using the following formula:

[0158] ;

[0159] in, Let t be the influent flow rate and pollutant load. As the baseline load, The amplitude of the diurnal cycle variation. This is the daily periodic term, used to simulate the periodic changes within a 24-hour day using a sine function. The phase angle, The amplitude of the periodic variation represents the degree of load difference between weekdays and weekends. This is a periodic term used to capture the systematic differences between weekend drainage patterns and weekday drainage patterns. The amplitude of seasonal variation characterizes the magnitude of load changes between different seasons. This is a seasonal term used to describe the pattern of load variation with the seasons. It represents a random fluctuation term.

[0160] Specifically, when establishing influent water quality and quantity prediction models for short-term, medium-term, and long-term based on extracted features, different prediction models are used for different prediction scales:

[0161] Short-term forecasting uses a time series model, taking the inflow data from the previous few hours as input, to predict high-frequency fluctuations in the next few hours.

[0162] The medium-term forecast uses a regression model based on periodic decomposition, focusing on predicting daily variation patterns;

[0163] Long-term forecasts employ machine learning models that combine weather forecasts and calendar information to predict seasonal trends and the impact of special events.

[0164] Preferably, when real-time inflow data is matched with historical patterns to drive the prediction model to generate a predicted trajectory of future inflow load, and the uncertainty range of the prediction is quantified, the generated prediction result is in the form of a probability distribution, expressed as follows:

[0165] ;

[0166] in, To predict the mean, To predict the standard deviation, Here is the confidence coefficient. This is to form the prediction interval for the inflow load at time t in the future.

[0167] Specifically, when providing the inflow forecast results and their uncertainty range to the intelligent decision-making module to generate advance control instructions for future load changes, the generation of advance control instructions is based on the following principles:

[0168] When a high load shock is predicted, increase the reserve of aeration capacity and carbon source addition capacity in advance.

[0169] When low-load periods are predicted, reduce energy consumption in advance and operate in energy-saving mode;

[0170] When the uncertainty of prediction is high, a robust optimization strategy is adopted to ensure that the target can be achieved even in the worst case.

[0171] Preferably, the prediction and control module also includes prediction performance evaluation and model adaptive update steps during operation:

[0172] The actual water inflow data is continuously compared with the predicted values ​​to calculate the prediction error index. When the prediction error continues to exceed the threshold, the parameters of the prediction model are automatically updated or the structure is adjusted to achieve online self-learning of prediction capabilities.

[0173] In this embodiment, the present invention systematically identifies the periodic patterns of influent load across multiple time scales, including daily, weekly, and seasonal patterns, through a mathematical decomposition model system, and constructs short-term, medium-term, and long-term multi-scale prediction models accordingly. By matching real-time data with historical patterns, it generates a predicted trajectory and uncertainty range of future influent load expressed in the form of a probability distribution. This transforms the influent load from an uncontrollable external disturbance into a predictable system input, enabling the control system to adjust its operating strategy in advance based on the prediction results (such as increasing processing capacity before high loads and optimizing energy consumption during low loads), achieving a fundamental shift from passive response to proactive anticipation. By quantifying prediction uncertainty and supporting robust optimization, the system's resilience and stability against influent fluctuations are significantly enhanced. Combined with an online self-learning mechanism to continuously improve prediction accuracy, the invention ultimately provides crucial front-end decision support for the entire system's energy saving, stable compliance, and intelligent operation.

[0174] like Figure 2 As shown, this embodiment also provides a coupled control method for deep denitrification of wastewater, including the following steps:

[0175] S41: Key parameters in the biological treatment tank are obtained through real-time water quality monitoring, and standardized data are obtained after preprocessing.

[0176] S42: Based on standardized data, dynamically estimate microbial activity using fusion mechanisms and data-driven digital twin models;

[0177] S43: Combining the current system status with future influent predictions, generate prediction results of microbial status and effluent water quality through forward simulation;

[0178] S44: With multiple objectives of achieving effluent quality standards, minimizing energy consumption, and maximizing microbial activity, a rolling optimization algorithm is used to solve for the comprehensive control commands of aeration, carbon source addition, and environmental parameters.

[0179] S45: Execute control commands and perform closed-loop adjustment of equipment operating status;

[0180] S46: At the same time, based on the periodic patterns of historical water inflow data, future load fluctuations can be predicted, enabling advance adjustments to the control strategy.

[0181] Example 1: Synergistic Optimization Control for Coping with Low Temperatures and High Load Shocks in Winter

[0182] This embodiment demonstrates how the system can achieve stable compliance and energy saving when faced with a sudden influx of high-concentration water under low-temperature conditions in winter, through coordinated control of the entire chain.

[0183] As winter approaches, the water temperature at a wastewater treatment plant in northern China drops to 12 degrees Celsius. At this time, the system, through its real-time water quality monitoring module's sensor array, detects a slow upward trend in ammonia nitrogen concentration at the outlet of the aerobic zone of the biological reactor. The data acquisition and pretreatment unit, while transmitting standardized data, calculates the changes in the ammonia nitrogen degradation rate. This data is then transmitted in real-time to the microbial twin modeling module.

[0184] The module's mechanistic model library is based on the activated sludge model. Its mechanistic model sub-unit predicts a theoretical ammonia nitrogen degradation rate based on current water temperature, dissolved oxygen, and ammonia nitrogen concentrations. However, the data-driven residual model sub-unit, by learning from historical data, discovered that actual microbial activity is more severely inhibited at low temperatures than the model's theoretical value. The data assimilation engine, i.e., the ensemble Kalman filter algorithm, then begins to work. It fuses and dynamically corrects these minute real-time monitoring data points with the model predictions, ultimately outputting a real-time estimate of the maximum specific growth rate of nitrifying bacteria through a microbial state estimator. The digital twin model shows that the current nitrifying bacteria activity is already at the lower limit of the healthy range.

[0185] Meanwhile, the predictive control module, based on its influent prediction model, analyzes that a high-concentration wastewater peak from the industrial park will arrive within the next two hours. It sends this predicted influent load trajectory and its uncertainty range to the intelligent decision-making module in advance. The real-time prediction module then activates, performing forward simulation based on the corrected current system state (including reduced nitrifying bacteria activity) provided by the microbial twin modeling module, and incorporating the predicted future high-load influent data. The prediction result generator outputs a warning: If the current state continues, the effluent ammonia nitrogen will exceed the standard in two hours.

[0186] The intelligent decision-making module responded immediately, constructing a finite-time optimization problem with the core objectives of achieving effluent standards, minimizing energy consumption, and restoring microbial activity within a future timeframe. In the solution process, it not only optimized aeration rate and carbon source dosage but also innovatively incorporated water temperature as a co-optimization variable. By querying the built-in microbial activity-environment parameter response surface, it discovered that slightly increasing the water temperature to 14 degrees Celsius via a heat exchanger significantly stimulated nitrifying bacteria activity, with overall energy consumption lower than simply drastically increasing aeration rate. Under the premise of satisfying various equipment constraints, the optimization solver calculated a series of optimal control command sequences.

[0187] Therefore, the actuator module began to operate: the variable frequency blower slightly increased the aeration intensity in advance, the carbon source dosing pump prepared, and the heat exchanger began to gently raise the temperature of the mixed liquor. These measures were in place before the peak high-load wastewater flow arrived. When the high-load shock actually arrived, the microbial community had already recovered its activity in a more suitable environment, efficiently removing ammonia nitrogen, and ultimately achieving stable effluent quality that met standards. Throughout the process, the system also verified the effectiveness of the heating strategy through environmental parameter control efficiency evaluation, fine-tuned the response surface parameters, and completed online self-learning.

[0188] Example 2: Intelligent forward-looking control for achieving daily energy saving and resilient operation

[0189] This embodiment demonstrates how the system achieves a dynamic balance between energy saving and stable operation under daily conditions with large fluctuations in influent load through accurate prediction and multi-objective optimization.

[0190] In urban wastewater treatment plants where the quality and quantity of influent fluctuate significantly between day and night, this system operates continuously. The predictive control module deeply analyzes long-term historical data to accurately identify the strong periodic patterns of influent flow and pollutant load within a 24-hour period and between weekdays and weekends. It uses different models to generate probabilistic prediction trajectories of influent load for the next 24 hours, clearly indicating that the low-load period will occur from early morning to dawn, while two load peaks will appear in the morning and evening.

[0191] Before the low-load period, the real-time water quality monitoring module reported that all parameters were normal. The microbial twin modeling module estimated that microbial activity was at a good level. The real-time prediction module performed forward simulation based on the current state and predicted future low-load influent data, predicting that even if the operating intensity was reduced in the next few hours, the effluent would easily meet the standards.

[0192] Based on this optimistic prediction, the intelligent decision-making module adaptively adjusted the weights in the objective function when constructing the rolling optimization problem: while ensuring that the effluent quality tracking item met the standards, it significantly increased the weights of the microbial activity optimization item and the control penalty item. This means that the primary goal of optimization shifted from "ensuring compliance" to "energy saving, consumption reduction, and stable operation." The optimal control command sequence obtained by the solver instructed the actuator module: the variable frequency blower reduced its speed to decrease the aeration volume, and the carbon source dosing pump operated intermittently according to the minimum demand. The system smoothly entered a low-power mode, significantly saving energy.

[0193] When the midday load peak was predicted, the system took proactive measures. Upon receiving a warning from the prediction and control module, the intelligent decision-making module immediately adjusted and optimized the target weights, prioritizing ensuring the effluent met standards. It instructed the actuator module to slightly increase the aeration rate in advance and prepare for carbon source addition. Simultaneously, noticing a slight deviation of the pH sensor reading from the optimal range, it used a digital twin model simulation to discover that fine-tuning the pH could improve denitrification efficiency. Therefore, in the optimization solution, it used pH as a decision variable, generating a small amount of alkali solution addition instruction to maintain the biochemical environment in a state optimal for microbial metabolism.

[0194] When the actual peak load arrives, the system is already in optimal "ready-to-fight" condition. Thanks to proactive regulation, microbial activity is precisely maintained within its high-efficiency range, smoothly mitigating shocks and preventing water quality fluctuations. Throughout the process, the microbial twin modeling module continuously assimilates real-time data and updates model parameters; the predictive regulation module compares the actual influent with predicted values, constantly improving its prediction accuracy. All these modules are tightly coupled, enabling the system to function like an experienced expert, anticipating the future and precisely targeting its capabilities, ultimately achieving resilient, stable, and highly efficient operation under complex and variable conditions.

[0195] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A coupled control system for deep denitrification of wastewater, characterized in that: include: The real-time water quality monitoring module is used to collect water quality parameters in the biological treatment tank in real time. The microbial twin modeling module is used to build and dynamically update digital twin models that reflect microbial activity based on water quality parameters. The real-time prediction module is used to obtain the prediction results of the microbial state based on the digital twin model, current water inflow data and future water inflow prediction; The intelligent decision-making module is used to generate optimal control commands based on the predicted results of microbial status and digital twin models; The actuator module, connected to the intelligent decision module, is used to execute control commands and adjust the aeration equipment and carbon source dosing equipment. The prediction and control module is used to analyze historical influent data to predict the future changes in influent water quality and quantity, and to provide the prediction results to the intelligent decision-making module for advance control. The microbial twin modeling module specifically includes: The mechanism model library stores mathematical models that describe the biochemical reaction process of wastewater. These mathematical models contain key parameters that represent the physiological state of the microbial community. The hybrid model engine is used to build and run mechanism-data driven hybrid models, which consist of parallel mechanistic model sub-units and data-driven residual model sub-units. The data assimilation engine is used to receive real-time water quality data and use the data assimilation algorithm to fuse and dynamically correct the real-time water quality data with the mechanism model. The algorithm used by the data assimilation engine is the ensemble Kalman filter algorithm. Microbial state estimator, based on a calibrated model, outputs real-time estimates of microbial activity indicators that cannot be directly measured, thereby constructing and updating a digital twin that reflects the true state of microorganisms; The prediction process of the hybrid model engine is defined by the following formula: ； in, The predicted system state vector at time k includes measurable water quality concentration variables and indirect microbial physiological parameters. As a sub-unit of the mechanism model, Let k be the state vector at time k-1. The system input at time k includes the influent flow rate and influent water quality. This is the set of inherent parameters for the mechanistic model. This is a data-driven residual model subunit used to learn the predicted residuals of the mechanism model subunit. This refers to the system's historical operating data up to time k-1. These are adjustable parameters for the residual model sub-units; Mechanism Model Subunit The following kinetic equation correlates microbial activity with water quality changes: ； in, This refers to the substrate consumption or product formation rate. For the corresponding microbial physiological parameters, Microbial concentration, This is the microbial yield coefficient. Substrate concentration, It is the half-saturation constant. This is an environmental inhibitory factor item; Furthermore, by assimilating real-time measured water quality change rate data, the state vector is... and data-driven residual model subunit The outputs are jointly estimated; The real-time prediction module specifically includes: The prediction initialization unit is used to obtain the real-time corrected digital twin model and its current state vector from the microbial twin modeling module as the initial state for prediction. A forward simulation engine is used to drive the digital twin model to perform dynamic simulations along a future timeline; The prediction result generator is used to output predicted data on the microbial state and effluent water quality over a future period based on simulation results.

2. The coupled control system for deep denitrification of wastewater according to claim 1, characterized in that: The execution steps of the data assimilation engine are as follows: Prediction step: Based on the state vector of the previous moment, forward simulation is performed through the mechanism model to obtain the predicted value of the state vector at the current moment and its uncertainty; Update step: Use the real-time water quality monitoring data at the current moment as the observed value, compare it with the model prediction value, calculate the Kalman gain, and make the optimal correction to the state vector prediction value to obtain the most likely state vector estimate at the current moment; Among them, the adjustable parameters of the data-driven residual model subunit in the state vector of the ensemble Kalman filter algorithm are... As part of the state vector, it is estimated and updated together with the microbial physiological parameters, thereby enabling the online self-learning of the residual model subunit.

3. The coupled control system for deep denitrification of wastewater according to claim 2, characterized in that: The simulation process of the forward simulation engine is defined by the following formula: ； in, The model state vector at time k is provided by the microbial twin modeling module and includes microbial physiological parameters and water quality concentration. Forward simulation function of digital twin model, The input sequence is the sequence of future H time steps starting from the current time k, containing future influent water quality and quantity data provided by the prediction and control module. For the set of model parameters, The input sequence is a sequence of state vectors for the next H time steps, predicted based on information at time k. In addition to future water inflow data, the input sequence also includes the control command sequence generated by the intelligent decision-making module as a known input to predict the system response under different control strategies.

4. The coupled control system for deep denitrification of wastewater according to claim 3, characterized in that: When the intelligent decision-making module is working, it specifically includes: S11: Obtain the current state vector of the digital twin model that has been corrected in real time by the microbial twin modeling module, as well as the system state prediction sequence in the future prediction time domain provided by the real-time prediction module; S12: Based on the current state and the state prediction sequence, construct a finite-time optimization problem with the future control command sequence as the decision variable; S13: Solve the optimization problem to obtain the optimal control command sequence; S14: Apply the first instruction in the optimal control instruction sequence to the actuator; S15: In the next control cycle, based on the new system state measurement value, repeat steps S11 to S14.

5. The coupled control system for deep denitrification of wastewater according to claim 4, characterized in that: The intelligent decision-making module is also used to generate control instructions for microbial environmental parameters, including temperature and pH, based on the digital twin model, in order to optimize the microbial state. Specifically: S21: Based on the current microbial physiological parameters and target effluent quality obtained from the microbial twin modeling module, determine the optimal activity range of key functional microbial communities and the corresponding target values ​​of optimal environmental parameters. S22: Using a digital twin model, the quantitative impact of minute changes in environmental parameters on microbial activity and effluent quality is analyzed through simulation, and a sensitivity matrix for environmental parameter regulation is established. S23: Environmental parameters are used as decision variables and incorporated together with operating parameters into the model predictive control framework for collaborative optimization and solution, generating a comprehensive control command sequence; S24: Extract environmental parameter control instructions from the optimal control instruction sequence, verify whether they are within the safe operating range of the equipment, and finally send the verified instructions to the actuator module.

6. The coupled control system for deep denitrification of wastewater according to claim 5, characterized in that: When the predictive regulation module is working, it specifically includes: S31: Analyze long-term historical inflow data to identify the periodic variation patterns of inflow flow and water quality parameters, including daily, weekly, and seasonal cycles, and extract key features; S32: Based on the extracted features, establish influent water quality and quantity prediction models for short-term, medium-term and long-term water use respectively; S33: Match real-time water inflow data with historical patterns to drive the prediction model to generate a predicted trajectory of future water inflow load and quantify the range of uncertainty in the prediction. S34: Provide the inflow prediction results and their uncertainty range to the intelligent decision-making module to generate advance control instructions for future load changes.

7. A coupled control method for deep denitrification of wastewater, applied to the system described in any one of claims 1-6, characterized in that: Includes the following steps: S41: Key parameters in the biological treatment tank are obtained through real-time water quality monitoring, and standardized data are obtained after preprocessing. S42: Based on standardized data, dynamically estimate microbial activity using fusion mechanisms and data-driven digital twin models; S43: Combining the current system status with future influent predictions, generate prediction results of microbial status and effluent water quality through forward simulation; S44: With multiple objectives of achieving effluent quality standards, minimizing energy consumption, and maximizing microbial activity, a rolling optimization algorithm is used to solve for the comprehensive control commands of aeration, carbon source addition, and environmental parameters. S45: Execute control commands and perform closed-loop adjustment of equipment operating status; S46: At the same time, based on the periodic patterns of historical water inflow data, future load fluctuations can be predicted, enabling advance adjustments to the control strategy.

Images