Digital twinborn monitoring method and system for multi-process coupling operation of assembled sewage plant

Abstract

The invention relates to the technical field of computers, discloses a digital twinborn monitoring method and system for multi-process coupling operation of an assembly type sewage plant, and aims to solve the problem that a traditional model is difficult to capture multi-process coupling nonlinear dynamic characteristics and improve the intelligent optimization operation level of the sewage plant under complex water quality and water quantity changes. According to the method, real-time data are collected and preprocessed, a hybrid digital twinborn model fusing a mechanism and data driving is constructed, state synchronization, prediction simulation, intelligent optimization decision and fault diagnosis early warning are carried out, and a control instruction is output. The system comprises a data acquisition preprocessing module, a digital twin modeling module, a simulation prediction module, an intelligent monitoring diagnosis module, an optimization decision module, a control instruction output module, a man-machine interaction module and the like. Physical digital bidirectional high-precision interaction and hybrid modeling are realized, the prediction and diagnosis capability is remarkably improved, the operation cost is reduced, the effluent is ensured to reach the standard, and the intelligent management of the sewage plant is comprehensively improved.

Description

Technical Field

[0001] This invention belongs to the field of industrial control technology, specifically relating to a digital twin monitoring method and system for the coupled operation of multiple processes in a prefabricated wastewater treatment plant. Background Technology

[0002] Accelerated global urbanization and sustained industrial economic growth have made wastewater treatment a crucial link in environmental health and sustainable development, placing high demands on the construction speed, operational efficiency, and environmental adaptability of wastewater treatment facilities. Against this backdrop, prefabricated wastewater treatment plants, with their characteristics of standardization, modularity, short construction cycles, and compact land use, are becoming an important development direction in modern wastewater treatment facility construction. This technology improves construction efficiency through rapid on-site assembly of prefabricated components and provides flexible solutions for dealing with complex and variable water quality and quantities.

[0003] To meet stringent effluent standards and cope with diverse influent water quality impacts, prefabricated wastewater treatment plants commonly adopt a multi-process coupled operation mode, organically integrating various biological treatment processes. This coupled mode aims to optimize treatment efficiency and enhance system resilience through the synergistic effect of different process units. However, the complex material transformations, energy exchanges, and hydraulic connections between different process units result in highly nonlinear and uncertain dynamic behavior of the entire system. Traditional single-mechanism models or data-driven models are insufficient to comprehensively and accurately capture its coupling mechanisms and dynamic characteristics.

[0004] Current technologies for the operation and management of prefabricated wastewater treatment plants generally rely on experience-based operations or static rule-based automated control. This approach lacks real-time perception of the system's overall state, accurate prediction of future trends, and intelligent optimization of operational strategies. Especially when facing external disturbances such as combined sewer overflows or sudden changes in ambient temperature, effectively coordinating the collaborative operation of multiple processes to achieve stable and compliant emissions while minimizing energy and carbon consumption has become a core bottleneck restricting its intelligent development. Although digital twin technology has demonstrated advantages in many industrial sectors, a mature hybrid modeling and simulation method has yet to be developed for the specific scenario of prefabricated wastewater treatment plants, which possess complex multi-process coupling. This method is insufficient to overcome the limitations of a single model and effectively handle the strong coupling relationships between different process modules, thus hindering the full realization of the potential of prefabricated wastewater treatment plants in intelligent operation, fault diagnosis, and optimization decision-making, necessitating innovative solutions. Summary of the Invention

[0005] This invention provides a digital twin monitoring method and system for the coupled operation of multiple processes in a prefabricated wastewater treatment plant. It aims to address the complex and variable water quality and quantity, overcoming the limitations of traditional single models in comprehensively capturing the nonlinear dynamic characteristics of coupled multiple processes, and realizing a shift in wastewater treatment plant operation management from experience-based operation to intelligent optimization decision-making. This invention constructs a digital twin model of the physical wastewater treatment plant to perform real-time monitoring, accurate prediction, intelligent diagnosis, and optimized control of the coupled multi-process system, ensuring stable effluent compliance while reducing operating energy consumption and carbon source consumption.

[0006] This invention proposes a digital twin monitoring method for the multi-process coupled operation of a prefabricated wastewater treatment plant. Its core lies in establishing a multi-level, bidirectional data flow and information interaction mechanism between the physical wastewater treatment plant and the digital twin model. This method encompasses everything from real-time acquisition and high-quality preprocessing of physical operation data, to the precise construction and online calibration of the multi-process coupled digital twin model, to simulation of operational status and prediction of future trends based on this digital twin model, followed by intelligent diagnosis of operational anomalies and comprehensive performance evaluation, ultimately achieving intelligent optimization of operational strategies and generation of specific control commands.

[0007] As one embodiment of the present invention, the digital twin monitoring method for multi-process coupled operation of the prefabricated wastewater treatment plant includes the following steps: Acquire real-time operating data of the prefabricated wastewater treatment plant, including environmental parameters, process parameters, and actuator status data collected by sensors; The real-time running data is preprocessed, including data cleaning, missing value imputation, outlier detection, and timestamp alignment. A multi-process coupled digital twin model of a prefabricated wastewater treatment plant is constructed. The multi-process coupled digital twin model includes a mechanism model module, a data-driven model module, and a multi-process coupled hybrid model construction module. The multi-process coupled digital twin model is used for real-time state synchronization, prediction and simulation, and intelligent optimization decision-making. Based on the results of the intelligent optimization decision, control commands are output to regulate the physical operation equipment of the prefabricated wastewater treatment plant; Based on the discrepancy between the real-time operating data and the predicted data from the digital twin model, fault diagnosis and early warning are performed.

[0008] Furthermore, the acquisition of real-time operational data for the prefabricated wastewater treatment plant specifically includes: real-time collection of water quality parameters, hydraulic parameters, operating environment parameters, and sludge characteristic parameters using high-precision online sensors deployed at various treatment units and key nodes of the prefabricated wastewater treatment plant. The water quality parameters include chemical oxygen demand (COD), biochemical oxygen demand (BOD), total nitrogen, total phosphorus, ammonia nitrogen, nitrate nitrogen, dissolved oxygen, oxidation-reduction potential, and pH. The hydraulic parameters include influent flow rate, effluent flow rate, liquid levels in each reaction tank, and sludge return flow rate. The operating environment parameters include ambient temperature and water temperature. The sludge characteristic parameters include mixed liquor suspended solids concentration, sludge settling ratio, and sludge volume index. Simultaneously, the status data of the actuators in the prefabricated wastewater treatment plant is acquired, including aeration fan speed, return pump start / stop status, dosing pump dosage, and valve opening. The data acquisition frequency is set to a preset periodic interval, for example, once every five minutes. The real-time operating data is uploaded to the data acquisition server via industrial Ethernet protocol or wireless communication protocol through programmable logic controller or distributed control system.

[0009] Furthermore, the preprocessing operation of the real-time running data specifically includes: Perform data cleaning to remove redundant information and erroneous records from the original data; Missing value imputation is performed. For data loss caused by sensor failure or transmission interruption, time series interpolation methods are used to supplement it. The time series interpolation methods include linear interpolation, Lagrange interpolation, or cubic spline interpolation. Perform outlier detection by identifying and correcting data points that significantly deviate from the normal range using statistical or machine learning methods. The statistical methods include the three-standard-deviation method and the box plot method, and the machine learning methods include the isolated forest algorithm and the local outlier algorithm. Perform timestamp alignment to ensure that data from different sensors and actuators are accurately synchronized in the time dimension.

[0010] Furthermore, the preprocessing operation also includes data standardization, which scales the preprocessed data to a preset numerical range or makes its mean zero and variance one, in order to eliminate the influence of different units of measurement on subsequent model training and analysis. The data standardization employs either the min-max normalization method or the Z-score standardization method.

[0011] Furthermore, the construction of the multi-process coupled digital twin model of the prefabricated wastewater treatment plant specifically includes: A mechanistic model module is constructed, comprising biochemical reaction kinetic models, hydraulic models, and mass-energy balance models for each treatment unit of the prefabricated wastewater treatment plant. This module covers enhanced anaerobic-aerobic process models, biological contact oxidation process models, aerobic granular sludge process models, and photosynthetic bacteria treatment process models. Each process model is custom-developed based on the activated sludge model framework, including key components such as dissolved matrix, granular matrix, dissolved oxygen, nitrogen compounds, phosphorus compounds, activated sludge, and inert sludge, as well as corresponding biokinetic, stoichiometric, and hydraulic parameters. The model parameters are calibrated offline using historical operating data and corrected in real-time using an online adaptive algorithm. A data-driven model module is constructed, which uses historical operational data to train a deep learning model to capture nonlinear and time-dependent features that are difficult for mechanistic models to describe. The deep learning model employs a Long Short-Term Memory (LSTM) network, a gated recurrent unit (GRU), or a transformer model based on an attention mechanism. The inputs to the deep learning model include preprocessed sensor data, actuator state data, and external disturbance data, including influent water quality prediction data, influent flow rate prediction data, and ambient temperature data. The outputs of the deep learning model include key water quality indicators, sludge characteristics, and energy consumption predictions. The deep learning model is trained using a gradient descent optimization algorithm, with mean squared error as the loss function. A multi-process coupled hybrid model is constructed, which integrates the mechanistic model module and the data-driven model module, and clarifies the material flow, energy flow, and hydraulic connectivity relationships between different process modules. The multi-process coupled hybrid model uses a directed graph structure to represent the process flow of the prefabricated wastewater treatment plant. The nodes of the directed graph represent individual treatment units, and the edges represent material exchange paths between units. The multi-process coupled hybrid model uses a Kalman filter or an extended Kalman filter to weight and fuse the prediction results of the mechanistic model and the data-driven model, and uses sensor observation data in real time for model state updates and error correction. The weighting coefficients are dynamically adjusted based on the prediction error covariance matrix of the mechanistic model and the data-driven model. The multi-process coupled hybrid model adaptively updates key model parameters based on the deviation between real-time operating data and the output of the digital twin model using recursive least squares or particle swarm optimization algorithms.

[0012] Furthermore, the use of the multi-process coupled digital twin model for real-time state synchronization, prediction and simulation, and intelligent optimization decision-making specifically includes: Real-time state synchronization is performed to accurately map the real-time operating data of the prefabricated wastewater treatment plant to the corresponding state variables of the digital twin model, ensuring that the physical entity and the digital twin are consistent in state. This real-time state synchronization employs an unscented Kalman filter or particle filter, fusing sensor observation data with prediction data from a multi-process coupled digital twin model to achieve accurate estimation of the system state. The system performs prediction and simulation, based on the current synchronized digital twin model state, to simulate the operational trends of the prefabricated wastewater treatment plant under different influent loads, environmental conditions, and control strategies over a future period. The prediction and simulation includes multi-step predictions of key indicators such as effluent quality, energy consumption, carbon source consumption, and sludge production. The system executes intelligent optimization decisions, generating optimal operation control strategies based on prediction and simulation results, combined with preset optimization objectives and constraints. The optimization objectives include achieving effluent discharge standards, minimizing energy consumption, minimizing carbon source consumption, and minimizing sludge production. The constraints include actuator physical limitations and process operating range. The intelligent optimization decision-making employs a model predictive control algorithm, combined with a genetic algorithm or particle swarm optimization algorithm, to iteratively solve for the future sequence of control variables to obtain the optimal control command. The control variables include aeration rate, return ratio, chemical dosage, and sludge discharge rate.

[0013] Furthermore, based on the results of the intelligent optimization decision, outputting control commands to regulate the physical operating equipment of the prefabricated wastewater treatment plant specifically includes: sending the control commands generated by the intelligent optimization decision to the actuators of the prefabricated wastewater treatment plant, such as aerators, return pumps, dosing pumps, and electric valves, through a distributed control system or programmable logic controller, to achieve precise adjustment of process parameters.

[0014] Furthermore, based on the discrepancy between the real-time operational data and the predicted data from the digital twin model, fault diagnosis and early warning specifically include: Continuously monitor the deviation between the real-time operating data of the prefabricated wastewater treatment plant and the predicted data of the multi-process coupled digital twin model; When the deviation exceeds a preset threshold, an anomaly detection is triggered. The anomaly detection employs a statistical process control method or a deep learning anomaly detection algorithm. Root cause analysis is performed on detected abnormal events. The root cause analysis uses knowledge graph-based reasoning methods or causal inference methods to identify the source and type of the fault. The fault type includes sensor fault, equipment fault, or process abnormality. Based on the fault diagnosis results, warning information is issued to operators through visual interfaces, audible and visual alarms, SMS or email, and fault handling suggestions are provided.

[0015] This invention also provides a digital twin monitoring system for the coupled operation of multiple processes in a prefabricated wastewater treatment plant, aiming to provide robust hardware and software support for the aforementioned methods. This system achieves comprehensive connectivity, real-time interaction, and intelligent decision-making between the physical wastewater treatment plant and the digital twin model.

[0016] As one embodiment of the present invention, the digital twin monitoring system for multi-process coupled operation of the prefabricated wastewater treatment plant includes: The data acquisition and preprocessing module is used to acquire multi-source heterogeneous operational data from physical wastewater treatment plants and perform cleaning, conversion and standardization processing. The digital twin modeling module is used to build and calibrate hybrid digital twin models with multiple coupled processes online. The simulation and prediction module is used for simulation analysis of operational status and prediction of future trends based on digital twin models. The intelligent monitoring and diagnostic module is used to monitor the operation of the wastewater treatment plant in real time, identify anomalies, and perform fault diagnosis and performance evaluation. The optimization decision-making module is used to generate optimized operating strategies and parameter recommendations based on monitoring, diagnosis, and prediction results. The control command output module is used to convert optimization strategies into executable control commands and send them to the physical wastewater treatment plant; The human-computer interaction interface module is used to provide a platform for user operation and information display.

[0017] Furthermore, the data acquisition and preprocessing module serves as the first barrier connecting the physical and digital worlds. This module includes a data interface unit, a data cleaning unit, and a feature engineering unit. The data interface unit supports various industrial communication protocols through different industrial communication interfaces, enabling data interaction with online sensors in the physical wastewater treatment plant, industrial automation control systems, and laboratory information management systems to acquire various operational data in real time. The industrial communication interfaces include serial interfaces (e.g., 485) and Ethernet interfaces. The industrial communication protocols include an open platform communication unified architecture, industrial Ethernet protocol, and message queue telemetry transmission protocol. The online sensors include water quality sensors, flow meters, level gauges, temperature sensors, and pressure sensors. The industrial automation control system includes distributed control systems and programmable logic controller systems. After receiving the raw data, the data cleaning unit performs outlier detection, missing value imputation, data denoising, and timestamp alignment to ensure data integrity, accuracy, and consistency. The feature engineering unit extracts temporal features and performs data standardization or normalization on the cleaned data. The temporal features include mean, variance, trend, and periodicity features. This processing generates a structured feature vector suitable for model input. All processed data is stored in a high-performance time-series database to support fast querying and historical backtracking.

[0018] Furthermore, the digital twin modeling module is the core intelligent engine of the entire system. This module includes a mechanism model library, a data-driven model library, a model integration unit, a parameter calibration unit, and a data assimilation unit. The mechanism model library stores detailed mathematical mechanism models for each process unit in the prefabricated wastewater treatment plant, including enhanced anaerobic-aerobic processes, biological contact oxidation processes, aerobic granular sludge processes, and photosynthetic bacteria treatment processes. The mathematical mechanism models include biochemical reaction kinetic models, hydraulic models, and mass and heat transfer models. The data-driven model library stores pre-trained machine learning and deep learning models, including long short-term memory networks and temporal convolutional networks, used to handle complex nonlinear processes that are difficult to accurately describe by mechanism models. The model integration unit is responsible for seamlessly coupling the heterogeneous models, achieving data exchange through material flow, energy flow, and information flow interfaces, and constructing a unified multi-process coupled digital twin model. The parameter calibration unit uses historical operating data and optimization algorithms to calibrate key parameters in the mechanism models offline or online. The data assimilation unit uses the Kalman filter algorithm to correct the state variables and parameters of the digital twin model online using real-time acquired data, ensuring the synchronization and high accuracy between the model and the physical entity.

[0019] Furthermore, the simulation and prediction module is responsible for extrapolating future operating states using a digital twin model. This module includes an operating condition simulator and a prediction engine. The operating condition simulator allows users or the system to define and simulate various operating scenarios, including changes in influent water quality and quantity, equipment failures, and drastic changes in environmental conditions. The simulator assesses the impact of these scenarios on the wastewater treatment plant's performance and generates detailed simulation reports. The prediction engine, based on the current state of the digital twin model and combined with historical data and future trends, performs short-term and long-term operating state predictions. The short-term prediction period includes several hours to several days, and the long-term prediction period includes several weeks to several months. The key prediction indicators include the dynamic trends of effluent water quality, energy consumption, and sludge production rate.

[0020] Furthermore, the intelligent monitoring and diagnostic module provides real-time operational status awareness and problem localization capabilities. This module includes an anomaly detection unit, a fault diagnosis engine, and a performance evaluation unit. The anomaly detection unit continuously monitors the deviation between the real-time operational data of the physical wastewater treatment plant and the predicted values ​​of the digital twin model, and uses various anomaly detection algorithms to identify various operational anomalies. These anomaly detection algorithms include unique class support vector machines and isolated forest algorithms. The operational anomalies include water quality exceeding standards, abnormal energy consumption, and equipment failure. The fault diagnosis engine combines a preset fault knowledge graph, expert rules, and causal reasoning mechanisms to conduct in-depth analysis of detected anomalies, locate the fault source, and provide root cause analysis. The performance evaluation unit, based on a multi-index evaluation system, quantitatively evaluates the wastewater treatment plant's operational efficiency, environmental benefits, economic benefits, and stability, generating a comprehensive performance score.

[0021] Furthermore, the optimization decision-making module is crucial for achieving intelligent operation. This module includes a target optimization engine and a strategy generator. The target optimization engine receives performance evaluation results from the intelligent monitoring and diagnosis module and prediction information from the simulation and prediction module. Combined with user-defined operational goals, including achieving effluent standards, minimizing energy consumption, and minimizing carbon source consumption, it uses a multi-objective optimization algorithm to iteratively optimize within a digital twin model. This multi-objective optimization algorithm includes a non-dominated sorting genetic algorithm type II and a reinforcement learning algorithm, generating a series of optimal operational strategies that satisfy the constraints. The strategy generator transforms the optimization results into intelligent recommended values ​​for key process parameters such as aeration rate, reflux ratio, sludge discharge rate, and reagent dosage, considering the synergistic effects between multiple processes to ensure the global optimality of the recommended strategies.

[0022] Furthermore, the control command output module is responsible for translating intelligent decisions into physical actions. This module includes an command conversion unit and a communication interface unit. The command conversion unit converts the high-level strategies generated by the optimization decision module into control commands that can be directly executed by the underlying automation system of the physical wastewater treatment plant. The underlying automation system of the physical wastewater treatment plant includes a programmable logic controller and a distributed control system. The communication interface unit establishes a connection with the control system of the physical wastewater treatment plant via industrial Ethernet and securely and reliably sends control commands to the field actuators in real time according to standard industrial communication protocols. The standard industrial communication protocols include an open platform communication unified architecture and the industrial Ethernet protocol. Simultaneously, this module is also responsible for receiving execution feedback from the control commands, forming a closed-loop control.

[0023] Furthermore, the human-machine interface module provides an intuitive user operation and information display platform. This module includes a data visualization component, an alarm management component, and a strategy configuration component. The data visualization component displays real-time operational data of the physical wastewater treatment plant, simulation results of the digital twin model, predicted trends, and dynamic changes of key indicators in the form of charts, dashboards, etc. The alarm management component centrally displays all early warning and fault diagnosis information and provides alarm response and confirmation functions. The strategy configuration component allows operators to configure operational goals, adjust the weights of optimization strategies, and view, approve, or manually intervene in the control commands recommended by the system. Users can achieve comprehensive management and control of the entire digital twin monitoring system through this module.

[0024] Compared with the prior art, the advantages and positive effects of the present invention are as follows: This invention constructs a comprehensive digital twin system, realizing bidirectional mapping and real-time interaction between the physical entity of a prefabricated wastewater treatment plant and its digital model. The system not only senses the real-time operating status of the physical wastewater treatment plant but also feeds back the plant's operational data to the digital twin model, enabling continuous optimization and correction of the model and ensuring a high degree of fit and accuracy between the digital twin model and the physical plant.

[0025] This invention proposes and implements for the first time a multi-process coupled hybrid modeling method, which deeply integrates the physicochemical principles of traditional mechanistic models with the nonlinear learning capabilities of data-driven models. The hybrid model overcomes the limitations of single mechanistic models in capturing the dynamic changes of complex biological processes and the poor interpretability of single data-driven models. It can more accurately describe the complex material transformation, energy exchange, and hydraulic connections between different biological treatment processes within prefabricated wastewater treatment plants, significantly improving the model's prediction accuracy and robustness.

[0026] This invention provides an intelligent optimization decision-making mechanism that, based on the accurate prediction and simulation capabilities of a digital twin model, can generate optimal operation control strategies for different operating conditions. These conditions include water quality and quantity shocks, and changes in ambient temperature. The optimization strategy comprehensively considers multiple objectives such as achieving effluent standards, minimizing energy consumption, minimizing carbon source consumption, and minimizing sludge production, significantly reducing the operating costs of wastewater treatment plants, improving treatment efficiency, and ensuring stable effluent quality.

[0027] This invention establishes a real-time fault diagnosis and early warning system. By continuously comparing the actual operating data of the physical plant with the predicted data of the digital twin model, it can quickly and accurately identify problems such as sensor failures, equipment anomalies, and process deviations. The system not only issues timely warnings but also performs root cause analysis and provides handling suggestions, significantly reducing unplanned downtime, ensuring the stable operation of the wastewater treatment plant, and avoiding environmental pollution risks.

[0028] This invention features high modularity and scalability. Each process module in the digital twin model is independent and can be flexibly combined and replaced through standardized interfaces, enabling it to adapt to prefabricated wastewater treatment plants of different configurations and scales, thus reducing the complexity of system deployment and maintenance. The system also boasts a high level of intelligence, reducing reliance on operator experience and enhancing the overall intelligent operation and management level of the prefabricated wastewater treatment plant. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the overall technical architecture of the digital twin monitoring system for multi-process coupled operation of prefabricated sewage treatment plants proposed in this invention.

[0030] Figure 2 This is a schematic diagram of the core principle framework of the multi-process coupled digital twin model in this invention.

[0031] Figure 3 This is a logical flowchart of the digital twin monitoring method for multi-process coupled operation of prefabricated sewage treatment plants in this invention.

[0032] Figure 4 This is a flowchart of the intelligent optimization decision-making process based on the digital twin model in this invention.

[0033] Figure 5 This is a schematic diagram of the multi-level interaction relationship and data flow between the physical wastewater treatment plant and the digital twin model in this invention. Detailed Implementation

[0034] This invention provides a digital twin monitoring method and system for the coupled operation of multiple processes in a prefabricated wastewater treatment plant. The aim is to construct a digital twin model of the physical wastewater treatment plant to enable real-time monitoring, accurate prediction, intelligent diagnosis, and optimized control of the coupled multi-process system, ensuring stable effluent quality while reducing energy consumption and carbon source consumption. This embodiment will elaborate on the method and its supporting system.

[0035] See Figure 3 The core of the digital twin monitoring method for multi-process coupled operation of prefabricated wastewater treatment plants lies in establishing a multi-level, bidirectional data flow and information interaction mechanism between the physical wastewater treatment plant and the digital twin model. This method encompasses everything from real-time acquisition and high-quality preprocessing of physical operation data, to the precise construction and online calibration of the multi-process coupled digital twin model, to the simulation of operational status and prediction of future trends based on this digital twin model, followed by intelligent diagnosis of operational anomalies and comprehensive performance evaluation, ultimately achieving intelligent optimization of operational strategies and generation of specific control commands. The method includes the following operational steps: The first step is to acquire real-time operating data of the prefabricated wastewater treatment plant. The real-time operating data includes environmental parameters, process parameters, and actuator status data collected by sensors.

[0036] In the operation of prefabricated wastewater treatment plants, the acquisition of real-time operational data is fundamental to the digital twin monitoring method. This step aims to comprehensively and accurately capture the instantaneous state of the physical wastewater treatment plant. Its specific implementation includes the following refined sub-steps: Sub-step 1: Deployment and data acquisition of high-precision online sensors.

[0037] By deploying a series of high-precision online sensors at each core treatment unit and key process node of the prefabricated wastewater treatment plant, continuous monitoring of multi-dimensional operating parameters is achieved. These sensors include, but are not limited to, chemical oxygen demand (COD) analyzers, biochemical oxygen demand (BOD) analyzers, total nitrogen (TN) analyzers, total phosphorus (TP) analyzers, ammonia nitrogen analyzers, nitrate nitrogen analyzers, dissolved oxygen sensors, oxidation-reduction potential (ORP) sensors, and pH meters for water quality parameter monitoring. Hydraulic parameters are collected through flow meters and level gauges installed at the inlet, outlet, and between each reaction tank, including influent flow rate, effluent flow rate, liquid level in each reaction tank, and sludge return flow rate. Operating environment parameters are collected through temperature sensors, including ambient temperature and water temperature. Sludge characteristic parameters are collected through online sludge concentration meters and sludge settling ratio analyzers, including mixed liquor suspended solids concentration, sludge settling ratio, and sludge volume index. Each sensor undergoes a rigorous calibration process before being put into operation to ensure its measurement accuracy meets industry standards. For example, the measurement error of the water quality sensor is controlled within 5%, and the measurement error of the level gauge is controlled at the millimeter level. The calibration cycle is set at once per quarter, and routine maintenance and cleaning are performed to avoid measurement drift caused by biofilm adhesion or impurities.

[0038] Sub-step 2: Acquiring actuator status data.

[0039] Besides process parameters, actuator status data is also a crucial component of the operational status of physical wastewater treatment plants. By interfaceing with the automated control system of the prefabricated wastewater treatment plant, the operating status of various actuators can be acquired in real time. This actuator status data includes the speed or frequency of aeration blowers, the start / stop status of return pumps, the actual dosage of chemicals by dosing pumps, and the opening degree of various electric valves. This data reflects the direct adjustment actions of the control system to process parameters and is key to understanding the system's operating logic and diagnosing faults. Actuator status data is read through its corresponding programmable logic controller (PLC) or distributed control system interface.

[0040] Sub-step 3: Setting the data acquisition frequency and transmission protocol.

[0041] To ensure the real-time nature and validity of the data, the data acquisition frequency is set to a preset periodic interval, such as once every five minutes. For parameters with higher response speed requirements, such as dissolved oxygen and aeration fan speed, the acquisition frequency can be increased to once per minute. All real-time operating data is initially aggregated and encapsulated by a field programmable logic controller or distributed control system via industrial Ethernet protocol or wireless communication protocol, and then uploaded to a central data acquisition server. Industrial Ethernet protocol ensures the bandwidth and stability of data transmission, while wireless communication protocol provides support for the flexibility of sensor deployment, especially suitable for renovation projects or areas where physical wiring is difficult to lay. During data transmission, a data verification mechanism, such as cyclic redundancy check, is used to ensure the integrity and accuracy of the data during transmission. The data packet structure includes a timestamp, sensor identifier, parameter type, measurement value, and unit, and follows a unified data format standard for subsequent processing.

[0042] The second step is to perform preprocessing operations on the real-time running data, including data cleaning, missing value imputation, outlier detection, and timestamp alignment.

[0043] Real-time data is inevitably affected by various noises, interferences, or equipment malfunctions during the acquisition process, resulting in poor data quality. Therefore, high-quality data preprocessing is a prerequisite for the accurate operation of the subsequent digital twin model. This step aims to transform the raw, messy data into a clean, complete, consistent, and high-quality dataset suitable for model input. Its specific implementation includes the following refined sub-steps: Sub-step 1: Data cleaning.

[0044] The goal of this sub-step is to remove redundant information and erroneous records from the raw data. Redundant information includes duplicate sensor readings, invalid sensor identifiers, or background data irrelevant to the operation of the wastewater treatment plant. Erroneous records mainly refer to outrageous data that does not conform to data format specifications or whose values ​​exceed the physical or process-permissible range. For example, a flow meter suddenly outputs a negative value during operation, or the dissolved oxygen concentration still shows an excessively high value after aeration has stopped. Data cleaning is achieved through preset business rules and logical checks. For example, a threshold check is set for the influent flow rate ranging from zero to one million cubic meters per day, and a range check is set for pH ranging from zero to fourteen. Any data points outside these ranges will be marked as erroneous records and processed according to preset strategies, such as direct deletion or marking as missing values ​​to be filled.

[0045] Sub-step 2: Fill in missing values.

[0046] For missing operational data due to sensor failure, communication interruption, or equipment maintenance, effective supplementation is necessary to ensure the integrity of the data sequence. This sub-step employs various time series interpolation methods for supplementation, including linear interpolation, Lagrange interpolation, and cubic spline interpolation. Linear interpolation is suitable for situations where the missing data time window is short and the data change trend is relatively stable, estimating the missing value by connecting adjacent data points with straight lines. Lagrange interpolation fits the data by constructing a high-order polynomial, suitable for situations where there are more complex nonlinear relationships between data points, but it is sensitive to the number and distribution of data points. Cubic spline interpolation fits the data using piecewise low-order polynomials, offering better smoothness and local adaptability, and is a commonly used method for handling medium-length missing sequences. When selecting a specific interpolation method, the system comprehensively considers the duration of the missing data, the change trend of the preceding and following data, and the impact on model accuracy. For example, for short-term missing dissolved oxygen data, due to its strong volatility, cubic spline interpolation can be preferentially used to capture its dynamic changes; for long-term missing liquid level data, linear interpolation may meet basic requirements if there are no major process adjustments. The populated data will be cross-validated with the original data to ensure its reasonableness.

[0047] Sub-step 3: Outlier detection.

[0048] Data points that significantly deviate from the normal range are identified and corrected using statistical or machine learning methods. These data points may be caused by sensor drift, transient interference, or abnormal operating conditions, but are not actual faults. The statistical methods include the three-standard-deviation method and box plots. The three-standard-deviation method calculates the mean and standard deviation of the data sequence and identifies data points exceeding three standard deviations of the mean as outliers. The box plot method defines the outlier determination boundary using the quartiles and interquartile ranges of the data, and is more suitable for non-normally distributed data. The machine learning methods include the Isolation Forest algorithm and the Local Anomaly Factor algorithm. The Isolation Forest algorithm isolates outliers more easily by randomly selecting features and split points, with relatively short path lengths. The Local Anomaly Factor algorithm identifies anomalies by calculating the local density of a data point relative to its neighbors, and is suitable for datasets with uneven density. After detecting outliers, the system corrects them according to a preset strategy, such as replacing them with historical means or medians, or using interpolation results from adjacent time points, rather than simply deleting them. The threshold and model parameters for outlier detection are regularly optimized and adjusted based on historical operating data to adapt to changes in the operating mode of the wastewater treatment plant.

[0049] Sub-step four: Timestamp alignment.

[0050] Prefabricated wastewater treatment plants deploy a large number of heterogeneous sensors and actuators, which typically have different sampling frequencies and data transmission delays. Timestamp alignment ensures that data from different sensors and actuators are precisely synchronized in the time dimension, providing unified time-series data for subsequent model input. This sub-step achieves data alignment by unifying all data sampling frequencies to a preset standard frequency, such as once per minute or once every five minutes. When the original data sampling frequency is higher than the standard frequency, downsampling methods are used, such as averaging or taking the last value. When the original data sampling frequency is lower than the standard frequency, upsampling methods are used, such as zero-order hold or linear interpolation. The timestamps of all data are converted to a unified Coordinated Universal Time (UTC) and strictly synchronized at the nanosecond level. Any data with a timestamp deviation exceeding a preset threshold is marked and corrected or discarded to avoid data misalignment and model misjudgment due to time asynchrony.

[0051] Sub-step five: Data standardization.

[0052] The preprocessing operation also includes data standardization, which scales the preprocessed data to a preset numerical range or makes its mean zero and variance one, to eliminate the influence of different units and numerical ranges on subsequent model training and analysis, and to ensure fair weighting of various data types in the model. Data standardization employs either min-max normalization or Z-score standardization. Min-max normalization linearly scales the data to a specific range, such as zero to one, suitable for situations where the data distribution range is known and there are no extreme outliers. Z-score standardization converts the data into a standard normal distribution with a mean of zero and a variance of one, suitable for situations where the data distribution is unknown or contains outliers, as it is insensitive to outliers. Standardized data can accelerate convergence during model training and prevent certain features from dominating model decisions due to excessively large values. For example, the numerical range of influent flow rate may be much larger than that of dissolved oxygen concentration; without standardization, flow rate features may have excessively high weights in the model.

[0053] Step 3, see Figure 2 A multi-process coupled digital twin model of a prefabricated wastewater treatment plant is constructed. The multi-process coupled digital twin model includes a mechanism model module, a data-driven model module, and a multi-process coupled hybrid model construction module.

[0054] Digital twin models are central to achieving intelligent monitoring and optimized decision-making. This step aims to construct a digital replica of the prefabricated wastewater treatment plant, accurately reflecting its dynamic behavior based on its physical characteristics and historical operational data. This model integrates the physicochemical principles of traditional mechanistic models with the nonlinear learning capabilities of data-driven models. Its implementation includes the following refined sub-steps: Sub-step 1: Construct the mechanism model module.

[0055] The mechanism model module includes biochemical reaction kinetic models, hydraulic models, and mass-energy balance models for each treatment unit of the prefabricated wastewater treatment plant. These models are custom-developed based on the activated sludge model framework and include key components such as dissolved matrix, granular matrix, dissolved oxygen, nitrogen compounds, phosphorus compounds, activated sludge, and inert sludge, as well as corresponding bioreaction kinetic parameters, stoichiometric parameters, and hydraulic parameters.

[0056] In detail, the mechanism model module includes enhanced anaerobic-aerobic process model, biological contact oxidation process model, aerobic granular sludge process model, and photosynthetic bacteria treatment process model.

[0057] An enhanced anaerobic-aerobic process model was constructed based on Activated Sludge Model No. 3 or its improved version. It details the biological reaction processes within the anaerobic, anoxic, and aerobic tanks, including substrate degradation, nitrification / denitrification, and phosphorus uptake and release by polyphosphate-accumulating bacteria, as well as the hydraulic processes such as sludge recirculation and mixed liquor recirculation between the tanks. The model includes biokinetic parameters such as maximum bacterial growth rate, substrate half-saturation constant, decay coefficient, and yield coefficient, as well as key parameters such as dissolved oxygen saturation constant, nitrification / denitrification rate constant, and phosphorus accumulation rate.

[0058] Biological contact oxidation process model: This model focuses on describing the microbial growth, substrate adsorption, and degradation processes on the biofilm. It considers factors such as biofilm thickness, mass diffusion resistance within the biofilm, and oxygen mass transfer rate, and is coupled with a suspended activated sludge model to comprehensively reflect the synergistic effect between the fixed biofilm and suspended organisms.

[0059] Aerobic granular sludge process model: Considering the unique structure of aerobic granular sludge, the model takes into account the mass transfer limitations within the granules, the effects of dissolved oxygen and substrate concentration gradients at different depths on the microbial reaction rate. The model includes the kinetic processes of granular sludge formation, growth, and stability maintenance, as well as the influence of granular structure on sludge-water separation performance.

[0060] A photosynthetic bacteria treatment process model describes the process by which photosynthetic bacteria utilize organic and inorganic matter for photosynthesis and heterotrophic metabolism under specific light and anaerobic conditions. The model considers the effects of light intensity, temperature, redox potential, and substrate concentration on the growth of photosynthetic bacteria and the efficiency of pollutant removal.

[0061] Each process model is custom-developed based on the activated sludge model framework to ensure the accuracy and versatility of the model in describing biological processes. The model parameters are calibrated offline using historical operating data. Offline calibration is achieved through global optimization algorithms, such as genetic algorithms or particle swarm optimization, to minimize the mean square error between model predictions and historical observations. The calibrated parameters are stored in a model parameter library. Furthermore, the model parameters are corrected in real-time using online adaptive algorithms, such as recursive least squares or extended Kalman filters, to fine-tune the model parameters using the latest real-time operating data, adapting to dynamic changes in operating conditions such as water quality, quantity, and ambient temperature, ensuring the model's real-time accuracy and robustness.

[0062] Sub-step 2: Construct the data-driven model module.

[0063] The data-driven model module uses historical operational data to train a deep learning model to capture nonlinear and temporally dependent features that are difficult for mechanistic models to describe. The deep learning model employs a long short-term memory network, a gated recurrent unit, or a transformer model based on an attention mechanism.

[0064] Long Short-Term Memory (LSTM) networks excel at processing and predicting time-series data, effectively learning long-term dependencies and avoiding gradient vanishing or exploding problems. Their internal input, forget, and output gates precisely control the inflow, forgetting, and output of information, enabling them to capture complex temporal characteristics in wastewater treatment processes. Input data includes continuous multi-day sensor sequence data and actuator operation records. Output is a prediction of key water quality indicators for the next few hours.

[0065] Gated recurrent units (GRUs) are simplified versions of long short-term memory networks. They have fewer parameters and faster training speeds, but can achieve similar performance on many time series tasks, making them suitable for real-time prediction scenarios with high computational resource requirements.

[0066] Transformer models based on attention mechanisms can process the entire sequence data in parallel and capture the dependency between any two positions in the sequence through a self-attention mechanism. They are particularly suitable for processing multivariate time series data with complex nonlinear relationships and long-distance dependencies.

[0067] The input to the deep learning model includes preprocessed sensor data, actuator status data, and external disturbance data. The sensor data covers all water quality, hydraulic, environmental, and sludge characteristic parameters. The actuator status data includes aeration fan speed, return pump start / stop status, dosing pump dosage, and valve opening. The external disturbance data includes influent water quality prediction data, influent flow rate prediction data, and ambient temperature data, which are typically obtained through external weather forecasts and influent water quality and flow rate prediction models. The output of the deep learning model includes key water quality indicators, sludge characteristics, and energy consumption predictions, such as effluent chemical oxygen demand (COD), ammonia nitrogen, total phosphorus concentration, mixed liquor suspended solids concentration, and total energy consumption predictions. The deep learning model is trained using a gradient descent optimization algorithm and employs mean squared error as the loss function to minimize the deviation between model predictions and actual observations. Batch standardization and dropout layers are used during training to prevent overfitting and improve the model's generalization ability.

[0068] Sub-step 3: Construct a multi-process coupled hybrid model.

[0069] The multi-process coupled hybrid model integrates the mechanistic model module and the data-driven model module, and clarifies the material flow, energy flow, and hydraulic connectivity relationships between different process modules. This fusion model aims to combine the physical interpretability of the mechanistic model with the nonlinear capture capability of the data-driven model, overcoming the limitations of a single model.

[0070] The multi-process coupled hybrid model uses a directed graph structure to represent the process flow of the prefabricated wastewater treatment plant. The nodes of the directed graph represent various treatment units, such as anaerobic tanks, anoxic tanks, aerobic tanks, secondary sedimentation tanks, sludge return pump stations, and chemical dosing systems. The edges of the directed graph represent the material exchange paths between units, such as influent flow, effluent flow, sludge return, nitrification liquor return, excess sludge discharge, and chemical dosing. Each edge includes attribute information such as material flow rate and concentration to ensure the accuracy of material balance calculations.

[0071] The multi-process coupled hybrid model uses a Kalman filter or an extended Kalman filter to weightedly fuse the predictions from the mechanistic model and the data-driven model, and utilizes sensor observation data in real time for model state updates and error correction. The Kalman filter is suitable for linear systems, while the extended Kalman filter extends the Kalman filter to nonlinear systems through linearization. During the fusion process, the predictions of each model are treated as observations, and the fusion algorithm dynamically adjusts the weighting coefficients based on their prediction error covariance matrices. For example, when the mechanistic model exhibits higher accuracy under specific operating conditions, its weight is increased accordingly; conversely, if the data-driven model is better at capturing instantaneous fluctuations, its weight is increased. This mechanism ensures that the model can adaptively utilize the advantages of each module.

[0072] The weighting coefficients are dynamically adjusted based on the prediction error covariance matrices of the mechanistic model and the data-driven model, where the error covariance matrix reflects the uncertainty of the model's prediction. The diagonal elements of the matrix represent the prediction variance of each state variable, while the off-diagonal elements represent the covariance between different state variables. By comparing the error covariance matrices output by the two models, the system can determine in real time which model's prediction of a specific state variable is more reliable at the current moment, thus assigning it a higher weight.

[0073] The multi-process coupled hybrid model adaptively updates key model parameters based on the discrepancy between real-time operating data and the output of the digital twin model, using either recursive least squares or particle swarm optimization (PSO) algorithms. Recursive least squares can estimate model parameters online and is suitable for systems where parameters change slowly over time. PSO, by simulating bird flock foraging behavior, searches for the optimal solution in the parameter space, possessing strong global search capabilities and being suitable for optimizing nonlinear parameters. These online parameter update mechanisms ensure that the digital twin model remains highly consistent with the actual operating conditions of the physical wastewater treatment plant.

[0074] See Figure 4 The fourth step involves using the multi-process coupled digital twin model for real-time state synchronization, prediction and simulation, and intelligent optimization decision-making.

[0075] The completed multi-process coupled digital twin model is not static; rather, it maintains dynamic synchronization with the physical entity through real-time data streams and executes a series of advanced intelligent functions to provide decision support for operations management. Its specific implementation includes the following refined sub-steps: Sub-step 1: Perform real-time state synchronization.

[0076] The real-time operational data of the prefabricated wastewater treatment plant is precisely mapped to the corresponding state variables in the digital twin model, ensuring consistency in state between the physical entity and the digital twin. This process is the core of the digital twin's "mirror" function. The real-time state synchronization employs an unscented Kalman filter or a particle filter, fusing sensor observation data with prediction data from a multi-process coupled digital twin model to achieve accurate estimation of the system state. The unscented Kalman filter, by selecting a set of deterministic sampling points, can more accurately handle state estimation problems in nonlinear systems, avoiding the errors caused by the linearization of the extended Kalman filter. The particle filter, on the other hand, uses a set of randomly sampled particles to represent the posterior probability distribution, making it particularly suitable for non-Gaussian and strongly nonlinear systems, and capable of estimating multimodal distributions, providing more comprehensive state information. By fusing data directly observed by sensors with state variables predicted internally by the digital twin model based on mechanisms and data-driven approaches, it is possible to accurately estimate internal states that cannot be directly measured or are too costly to measure, such as biomass and substrate concentrations in each reactor—key factors for understanding and controlling biological processes.

[0077] Sub-step 2: Perform prediction and simulation.

[0078] Based on the current synchronized digital twin model status, the operational trends of the prefabricated wastewater treatment plant under different influent loads, environmental conditions, and control strategies over a future period are simulated. This sub-step provides the ability to anticipate possible future scenarios. The prediction and simulation include multi-step predictions of key indicators such as effluent quality, energy consumption, carbon source consumption, and sludge production. The prediction timeframe is typically set to the next 24 to 72 hours to support short-term operational adjustments. Simulation allows operators or the system to define specific "hypothetical" scenarios, such as a sudden 50% increase in influent chemical oxygen demand (COD) or a 5-degree Celsius drop in ambient temperature over the next eight hours, assessing the impact of these changes on effluent quality compliance, aeration energy consumption, carbon source dosage, and sludge production. Simulation results are presented in the form of dynamic curves, numerical tables, and early warning reports, providing decision-makers with quantitative data. For example, simulation can assess whether the current aeration rate can guarantee compliance with effluent dissolved oxygen and ammonia nitrogen standards under specific influent shocks; if not, an early warning will be triggered and adjustments will be recommended.

[0079] Sub-step 3: Execute intelligent optimization decisions.

[0080] Based on the prediction and simulation results, and combined with the preset optimization objectives and constraints, an optimal operation control strategy is generated. This sub-step is crucial for the intelligent operation management of the digital twin monitoring system. The optimization objectives include achieving effluent discharge standards, minimizing energy consumption, minimizing carbon source consumption, and minimizing sludge production. These objectives can be integrated into a comprehensive optimization objective function through weighted summation, or a Pareto optimal solution can be sought among multiple objectives using a multi-objective optimization algorithm. The constraints include physical limitations of actuators, such as the maximum speed of the aeration blower and the maximum flow rate of the return pump, as well as the process operating range, such as the minimum dissolved oxygen concentration and maximum sludge concentration in the reaction tank.

[0081] The intelligent optimization decision-making employs a model predictive control algorithm, combined with a genetic algorithm or particle swarm optimization algorithm, to iteratively solve the future control variable sequence to obtain the optimal control command. The model predictive control algorithm predicts the system behavior over a future period using a digital twin model within each control cycle, then continuously optimizes the future control variable sequence, thereby achieving the optimal objective function while satisfying constraints. The genetic algorithm searches the candidate control strategy space by simulating natural selection and genetic mechanisms, gradually evolving to the optimal solution through selection, crossover, and mutation operations. The particle swarm optimization algorithm simulates bird flock foraging behavior, where each "particle" represents a potential control strategy, updating its position by tracking individual optimal solutions and the global optimal solution, ultimately converging to the optimal strategy. The control variables include aeration rate, return ratio, chemical dosage, and sludge discharge rate; the optimized values ​​of these variables constitute the final control command. For example, the system calculates how to allocate aeration rates at different stages to minimize total energy consumption while ensuring that effluent ammonia nitrogen meets standards.

[0082] The fifth step is to output control commands based on the results of the intelligent optimization decision to regulate the physical operation equipment of the prefabricated wastewater treatment plant.

[0083] The results of intelligent optimization decisions must be translated into executable physical commands to effectively regulate the prefabricated wastewater treatment plant. This step is crucial for achieving closed-loop control using digital twins. Its implementation includes the following refined sub-steps: Sub-step 1: Control instruction generation and conversion.

[0084] The abstract operational strategies generated by the intelligent optimization decisions, such as "adjusting the dissolved oxygen target value in the aerobic tank to 2.0 mg / L, adjusting the sludge return ratio in the secondary sedimentation tank to 80%, and increasing the carbon source dosage in the anaerobic tank by 10%", are converted into control instructions that the underlying automation system of the physical wastewater treatment plant can directly understand and execute. These instructions are typically digital or analog signals, such as the speed setpoint of the aeration blower, the start or stop signal of the return pump, the flow setpoint of the dosing pump, and the opening percentage of the electric valve. The instruction conversion unit maps the high-level strategy instructions into specific binary or hexadecimal control words according to the communication protocols and control methods of different actuators. For example, the 80% return ratio is converted into the corresponding analog output value in the programmable logic controller to control the frequency converter of the return pump.

[0085] Sub-step two: Instruction transmission and issuance.

[0086] The converted control commands are sent to actuators such as aerators, return pumps, dosing pumps, and electric valves in the prefabricated wastewater treatment plant via a distributed control system or programmable logic controller (PLC) as middleware. Data transmission employs standard industrial communication protocols, such as an open platform communication architecture or industrial Ethernet protocol, ensuring real-time performance, reliability, and security of command transmission. The communication interface unit establishes a stable communication link and uses data frame verification and command confirmation mechanisms to prevent command loss or errors during transmission. Command issuance is typically performed cyclically, updating the actuator settings every preset time interval, such as ten or thirty seconds.

[0087] Sub-step 3: Actuator response and process parameter adjustment.

[0088] Upon receiving a control command, the corresponding actuators immediately activate to precisely adjust process parameters. For example, the aeration blower adjusts its speed according to the new setpoint, thereby changing the aeration volume and affecting the dissolved oxygen concentration in the aerobic tank; the return pump adjusts the sludge return flow rate according to the new start / stop or flow command, affecting the sludge-water separation effect in the secondary sedimentation tank and the sludge concentration in the anaerobic-anoxic tank; the dosing pump adjusts the dosage of carbon source or phosphorus removal agent according to the new dosage command to meet the needs of denitrification or chemical phosphorus removal; and the electric valve adjusts its opening according to the command, controlling the hydraulic flow direction or flow rate. The feedback mechanism inside the actuator uploads its execution status in real time, forming a closed-loop control to ensure accurate execution of commands. For example, the blower speed sensor detects and reports the actual speed; if there is a deviation from the setpoint, the control system will make fine adjustments until the target is reached.

[0089] The sixth step is to perform fault diagnosis and early warning based on the deviation between the real-time operating data and the predicted data of the digital twin model.

[0090] By continuously comparing actual operational data from physical wastewater treatment plants with predicted data from digital twin models, anomalies and potential faults in system operation can be effectively identified, enabling proactive fault diagnosis and early warning. This step is crucial for ensuring the safe and stable operation of wastewater treatment plants and preventing environmental pollution and economic losses. Its specific implementation includes the following refined sub-steps: Sub-step 1: Continuously monitor deviations.

[0091] The system continuously monitors the deviation between the real-time operating data of the prefabricated wastewater treatment plant and the predicted data of the multi-process coupled digital twin model. Deviation calculation can employ various indicators, such as absolute error, relative error, root mean square error, or statistical distance. Monitoring is continuous, for example, calculating the deviation value of key parameters every second or minute. By establishing a historical deviation baseline, it can be determined whether the current deviation is within the normal fluctuation range. For example, for effluent chemical oxygen demand (COD), if the deviation between the actual measured value and the predicted value of the digital twin model exceeds a certain preset percentage for half an hour consecutively, it is considered a preliminary abnormal signal.

[0092] Sub-step 2: Trigger abnormal event detection.

[0093] When the deviation continuously exceeds a preset threshold, an anomaly detection is triggered. The preset threshold is typically determined based on statistical analysis of historical operating data, expert experience, or specific process requirements. The anomaly detection employs statistical process control methods or deep learning anomaly detection algorithms.

[0094] Statistical process control methods include Shewhart control charts, cumulative sum control charts, or exponentially weighted moving average control charts. These methods can identify statistically significant changes in a data series, such as mean shifts, trend changes, or increased volatility. For example, if the deviation of dissolved oxygen falls outside the control limits for three consecutive points on a Shewhart control chart, it is considered an anomaly.

[0095] Deep learning anomaly detection algorithms include unsupervised learning methods based on autoencoders or generative adversarial networks. These models learn the data features of normal operating patterns and are able to identify data points that deviate significantly from normal patterns. For example, an autoencoder attempts to reconstruct the input data; if the reconstruction error exceeds a threshold, the input data is considered anomaly.

[0096] The abnormal event detection system has a multi-level early warning mechanism. The first level may be a minor warning, prompting operators to pay attention; the second level is a serious warning, which may be accompanied by audible and visual alarms; the third level directly triggers the fault diagnosis process.

[0097] Sub-step 3: Perform root cause analysis on the detected abnormal events.

[0098] This sub-step aims to identify the source and type of the fault. The root cause analysis employs knowledge graph-based reasoning or causal inference methods.

[0099] The reasoning method based on knowledge graphs involves pre-constructing a knowledge graph that includes relationships between wastewater treatment processes, equipment components, sensors, control logic, and common failure modes. Nodes in the knowledge graph represent entities, and edges represent relationships between entities. When an anomaly is detected, the reasoning engine performs pathfinding and logical reasoning within the knowledge graph based on the anomaly's characteristics, the time of occurrence, and the involved sensors and actuators to identify the most likely source of the failure. For example, if dissolved oxygen in the aeration tank remains consistently low while the aeration blower speed is normal, the knowledge graph might point to aeration head blockage or excessive biological load.

[0100] Causal inference methods: Utilizing historical data to analyze the causal relationships between different parameters. For example, Granger causality tests or causal discovery algorithms based on structural equation modeling can be used to identify which variable changes directly lead to the current anomaly. For instance, by analyzing the causal relationship between influent chemical oxygen demand (COD) and dissolved oxygen in the aerobic tank, it can be determined whether the low dissolved oxygen level is due to influent load shocks or a malfunction in the aeration system itself.

[0101] The fault types include sensor malfunctions, equipment malfunctions, or process anomalies. Sensor malfunctions may manifest as data drift, data freezing, or outputting incorrect values. Equipment malfunctions may include aeration blower wear, return pump blockage, or valve jamming. Process anomalies may include inhibited nitrification, sludge bulking, or effluent quality exceeding standards.

[0102] Sub-step four: Issue a warning message and provide troubleshooting suggestions.

[0103] Based on the fault diagnosis results, early warning information is issued to operators through visual interfaces, audible and visual alarms, SMS, or email, along with fault handling suggestions. The visual interface displays abnormal data, diagnostic results, fault location, and affected area using intuitive charts and highlights. Audible and visual alarms provide immediate on-site alerts. SMS and email can send detailed warning information to the mobile devices or email addresses of relevant personnel. Warning information includes the fault type, time of occurrence, possible causes, affected process units, and the system's recommended initial handling measures. For example, if the diagnosis result is "dissolved oxygen sensor malfunction, data frozen at low values," the system will suggest "checking and calibrating or replacing the dissolved oxygen sensor." If the diagnosis result is "excessive influent organic load leading to a decrease in dissolved oxygen in the aerobic tank," the system will suggest "increasing aeration or reducing influent flow." These handling suggestions help operators respond quickly and take effective corrective measures to minimize the impact of the fault on wastewater treatment efficiency and operating costs.

[0104] The above details the implementation steps of the digital twin monitoring method for the coupled operation of multiple processes in a prefabricated wastewater treatment plant. To support the effective operation of this method, this invention also provides a digital twin monitoring system for the coupled operation of multiple processes in a prefabricated wastewater treatment plant. This system achieves comprehensive connection, real-time interaction, and intelligent decision-making between the physical wastewater treatment plant and the digital twin model.

[0105] See Figure 1 and Figure 5 The digital twin monitoring system for the multi-process coupled operation of the prefabricated wastewater treatment plant includes the following modules: Part 1: Data Acquisition and Preprocessing Module.

[0106] This module serves as the first barrier connecting the physical and digital worlds, responsible for acquiring multi-source, heterogeneous operational data from physical wastewater treatment plants and performing cleaning, transformation, and standardization. Internally, this module includes a data interface unit, a data cleaning unit, and a feature engineering unit.

[0107] Data Interface Unit: Supporting various industrial communication protocols through different industrial communication interfaces, this unit interacts with online sensors in the wastewater treatment plant, industrial automation control systems, and laboratory information management systems to acquire various operational data in real time. The industrial communication interfaces include a serial 485 interface and an Ethernet interface. The serial 485 interface is suitable for long-distance, multi-point communication, while the Ethernet interface provides high-bandwidth, high-speed data transmission capabilities. The industrial communication protocols include an open platform communication unified architecture, industrial Ethernet protocol, and message queue telemetry transmission protocol. The open platform communication unified architecture provides a unified data access interface and semantic model, enabling interoperability between devices from different manufacturers. The industrial Ethernet protocol ensures reliable transmission of field data. The message queue telemetry transmission protocol is suitable for lightweight data transmission in low-bandwidth, high-latency network environments. The online sensors include water quality sensors, flow meters, level gauges, temperature sensors, and pressure sensors. The industrial automation control system includes distributed control systems and programmable logic controller systems. The design of the data interface unit fully considers the communication characteristics and data formats of different devices, achieving seamless access to multi-source data.

[0108] The data cleaning unit receives raw data and performs operations such as outlier detection, missing value imputation, data denoising, and timestamp alignment to ensure data integrity, accuracy, and consistency. This unit integrates various data processing algorithms, such as statistical methods like three-standard-deviation and boxplots for outlier detection, time-series-based linear interpolation, Lagrange interpolation, and cubic spline interpolation for missing value imputation, and wavelet transform or moving average-based data denoising algorithms. Timestamp alignment is achieved through a high-precision clock synchronization service.

[0109] Feature engineering unit: This unit extracts time-series features and performs data standardization or normalization on the cleaned data. The time-series features include mean, variance, trend, and periodicity, such as calculating the mean and standard deviation within a sliding window, and extracting the Fourier transform spectrum of the data to identify periodic components. This processing generates a structured feature vector suitable for model input. Data standardization uses min-max normalization or Z-score normalization to scale the data to a uniform scale. All processed data is stored in a high-performance time-series database to support fast querying and historical backtracking. This database supports high-speed writing, compressed storage, and efficient querying, ensuring data traceability.

[0110] Part Two: Digital Twin Modeling Module.

[0111] This module is the core intelligent engine of the entire system, used to build and calibrate hybrid digital twin models with multiple coupled processes online. This module includes a mechanistic model library, a data-driven model library, a model integration unit, a parameter calibration unit, and a data assimilation unit.

[0112] Mechanism Model Library: This library stores detailed mathematical mechanism models for various process units in prefabricated wastewater treatment plants, including enhanced anaerobic-aerobic processes, biological contact oxidation processes, aerobic granular sludge processes, and photosynthetic bacteria treatment processes. The mathematical mechanism models include biochemical reaction kinetic models, hydraulic models, and mass and heat transfer models. These models are stored as modular components, each with clearly defined input and output interfaces and internal state variables.

[0113] Data-Driven Model Library: This library stores pre-trained machine learning and deep learning models, including Long Short-Term Memory networks and Temporal Convolutional Networks, designed to handle complex nonlinear processes that are difficult to accurately describe using mechanistic models. These models are deployed as reusable services and support real-time inference.

[0114] The model integration unit is responsible for seamlessly coupling the heterogeneous models, enabling data exchange through material flow, energy flow, and information flow interfaces, and constructing a unified multi-process coupled digital twin model. This unit adopts a service-oriented architecture and defines standardized interface protocols, allowing different types of models to work collaboratively. For example, the mechanistic model can provide predictions of specific state variables as input to the data-driven model, or the data-driven model can compensate for the residuals of the mechanistic model.

[0115] Parameter calibration unit: Utilizing historical operational data, this unit performs offline or online calibration of key parameters in the mechanistic model using optimization algorithms. It integrates multiple optimizers, including genetic algorithms, particle swarm optimization, and nonlinear least squares methods.

[0116] Data assimilation unit: Utilizing Kalman filtering algorithms, such as the extended Kalman filter or the unscented Kalman filter, this unit uses real-time acquired data to online correct the state variables and parameters of the digital twin model, ensuring synchronization and high accuracy between the model and the physical entity. This unit achieves real-time data and model fusion, continuously reducing model prediction errors.

[0117] Part Three: Simulation and Prediction Module.

[0118] This module is responsible for using digital twin models to predict future operating conditions. It includes a working condition simulator and a prediction engine.

[0119] Operating Condition Simulator: Allows users or the system to define and simulate various operating scenarios, including changes in influent water quality and quantity, equipment failures, and drastic changes in environmental conditions. The simulator assesses the impact of these scenarios on the wastewater treatment plant's performance and generates detailed simulation reports. Users can set simulation parameters through a graphical interface, such as setting a 25% increase in influent chemical oxygen demand (COD) over the next 24 hours. Based on a digital twin model, the simulator will extrapolate the system response over a simulation timeline, outputting dynamic curves and numerical reports for key indicators.

[0120] Prediction Engine: Based on the current state of the digital twin model, combined with historical data and future trends, it makes short-term and long-term operational status predictions. The short-term prediction period includes several hours to several days, and the long-term prediction period includes several weeks to several months. The key prediction indicators include the dynamic trends of effluent quality, energy consumption, and sludge production rate. The prediction engine utilizes algorithms such as ensemble learning and time series analysis, combined with the dynamic characteristics of the digital twin model, to output high-confidence future trend prediction results.

[0121] Part Four: Intelligent Monitoring and Diagnosis Module.

[0122] This module provides real-time operational status awareness and problem localization capabilities. It includes an anomaly detection unit, a fault diagnosis engine, and a performance evaluation unit.

[0123] Anomaly Detection Unit: This unit continuously monitors the deviation between real-time operational data of the physical wastewater treatment plant and the predicted values ​​of the digital twin model, and uses various anomaly detection algorithms to identify various operational anomalies. These algorithms include unique class support vector machines and isolated forest algorithms. The operational anomalies include water quality exceeding standards, abnormal energy consumption, and equipment malfunctions. This unit promptly identifies potential problems through multi-dimensional feature analysis and time series pattern matching.

[0124] Fault Diagnosis Engine: Combining a pre-defined fault knowledge graph, expert rules, and a causal reasoning mechanism, this engine performs in-depth analysis of detected anomalies, pinpointing the fault source and providing root cause analysis. The fault knowledge graph includes information such as process flow, equipment attributes, sensor locations, and historical fault cases, and is formally represented through ontology. The expert rule base stores diagnostic experience accumulated by industry experts. The causal reasoning mechanism reveals the underlying causes of anomalies by analyzing the causal chains between variables.

[0125] Performance Evaluation Unit: Based on a multi-indicator evaluation system, this unit quantitatively assesses the operational efficiency, environmental benefits, economic benefits, and stability of the wastewater treatment plant, generating a comprehensive performance score. Evaluation indicators include effluent quality compliance rate, energy consumption per unit of wastewater treatment, carbon source consumption, sludge production rate, equipment uptime, and maintenance costs. This unit provides multi-dimensional and multi-level performance reports, enabling operators to comprehensively understand the wastewater treatment plant's operational status.

[0126] Part Five: Optimization Decision Module.

[0127] This module is crucial for achieving intelligent operation. It includes a target optimization engine and a strategy generator.

[0128] The target optimization engine receives performance evaluation results from the intelligent monitoring and diagnosis module and prediction information from the simulation and prediction module. It then combines these with user-defined operational objectives, including achieving effluent standards, minimizing energy consumption, and minimizing carbon source consumption. A multi-objective optimization algorithm is used to iteratively optimize the digital twin model. This multi-objective optimization algorithm includes a non-dominated sorting genetic algorithm (NGS2) and a reinforcement learning algorithm. The NGS2 can simultaneously handle multiple optimization objectives and generate a Pareto optimal solution set, providing decision-makers with various trade-off options. The reinforcement learning algorithm learns the optimal control strategy through interaction with the digital twin model and is suitable for complex dynamic environments.

[0129] Strategy Generator: This generator transforms optimization results into intelligent recommended values ​​for key process parameters such as aeration rate, recirculation ratio, sludge discharge rate, and reagent dosage. It also considers the synergistic effects between multiple processes to ensure the global optimality of the recommended strategies. The strategy generator further verifies the feasibility of the recommended strategies, ensuring they meet physical constraints and process requirements, and generates detailed strategy reports.

[0130] Part 6, Control Command Output Module.

[0131] This module is responsible for translating intelligent decisions into physical actions. It includes an instruction conversion unit and a communication interface unit.

[0132] The instruction conversion unit converts the high-level strategies generated by the optimization decision module into control instructions that can be directly executed by the underlying automation system of the physical wastewater treatment plant. The underlying automation system includes a programmable logic controller (PLC) and a distributed control system. The instruction conversion unit has built-in instruction format and protocol converters for various actuator types.

[0133] Communication Interface Unit: This unit establishes a connection with the control system of the physical wastewater treatment plant via industrial Ethernet and securely and reliably sends control commands to field actuators in real time according to standard industrial communication protocols. These standard industrial communication protocols include an open platform communication unified architecture and the industrial Ethernet protocol. This unit is responsible for sending commands, transmission confirmation, and error handling. Simultaneously, this module is also responsible for receiving execution feedback from control commands, forming a closed-loop control system to ensure that commands are executed accurately.

[0134] Part 7, Human-Computer Interaction Interface Module.

[0135] This module provides an intuitive user interface and information display platform. It includes data visualization components, alarm management components, and policy configuration components.

[0136] Data visualization component: This component displays real-time operational data of the physical wastewater treatment plant, simulation results from the digital twin model, predicted trends, and dynamic changes in key indicators in the form of charts, dashboards, etc. It supports custom layouts and multi-screen displays, providing a comprehensive operational overview. For example, it can display influent and effluent water quality trend charts, real-time dissolved oxygen distribution maps, energy consumption curves, and predicted sludge production values.

[0137] Alarm Management Component: This component centrally displays all early warning and fault diagnosis information and provides alarm response and confirmation functions. It supports alarm classification, filtering by time or type, and alarm event tracing, assisting operators in efficiently handling anomalies.

[0138] The strategy configuration component allows operators to configure operational goals, adjust the weights of optimization strategies, and view, approve, or manually intervene in system-recommended control instructions. This component provides a flexible parameter setting interface and decision-making process management functions, enabling the system to operate autonomously while also allowing for human supervision and intervention, achieving intelligent management through human-machine collaboration. Users can achieve comprehensive management and control of the entire digital twin monitoring system through this module.

[0139] Through the above methods and systems, this invention realizes intelligent and refined operation and management of prefabricated wastewater treatment plants, overcomes the limitations of traditional single models in capturing the nonlinear dynamic characteristics of multi-process coupling, ensures stable effluent quality, and significantly reduces operating energy consumption and carbon source consumption, providing advanced technical support for the sustainable development of modern wastewater treatment plants.

[0140] The above are merely specific embodiments of the present invention, but the technical features of the present invention are not limited thereto. Any simple changes, equivalent substitutions, or modifications made based on the present invention to solve essentially the same technical problems and achieve essentially the same technical effects are all covered within the protection scope of the present invention.

Claims

1. A digital twin monitoring method for multi-process coupled operation of a prefabricated wastewater treatment plant, characterized in that, Includes the following steps: Step 1: Obtain the real-time operating data of the prefabricated wastewater treatment plant. The real-time operating data includes environmental parameters, process parameters, and actuator status data collected by sensors. Step 2: Perform preprocessing operations on the real-time running data, including data cleaning, missing value imputation, outlier detection, and timestamp alignment. Step 3: Construct a multi-process coupled digital twin model of the prefabricated wastewater treatment plant. The multi-process coupled digital twin model includes a mechanism model module, a data-driven model module, and a multi-process coupled hybrid model construction module. Step 4: Utilize the aforementioned multi-process coupled digital twin model for real-time state synchronization, prediction and simulation, and intelligent optimization decision-making; Step 5: Based on the results of the intelligent optimization decision, output control commands to regulate the physical operation equipment of the prefabricated wastewater treatment plant; and Step 6: Based on the deviation between the real-time operating data and the predicted data of the digital twin model, perform fault diagnosis and early warning.

2. The method according to claim 1, characterized in that, The acquisition of real-time operational data for prefabricated wastewater treatment plants specifically includes: High-precision online sensors deployed in various treatment units and key nodes of the prefabricated wastewater treatment plant are used to collect real-time data on water quality parameters, hydraulic parameters, operating environment parameters, and sludge characteristics. Acquire actuator status data of the prefabricated wastewater treatment plant, including aeration fan speed, return pump start / stop status, dosing pump dosage, and valve opening. The water quality parameters include chemical oxygen demand (COD), biochemical oxygen demand (BOD), total nitrogen, total phosphorus, ammonia nitrogen, nitrate nitrogen, dissolved oxygen, oxidation-reduction potential (ORP), and pH. The hydraulic parameters include influent flow rate, effluent flow rate, liquid level in each reaction tank, and sludge return flow rate. The operating environment parameters include ambient temperature and water temperature. The sludge characteristic parameters include mixed liquor suspended solids concentration, sludge settling ratio, and sludge volume index.

3. The method according to claim 1, characterized in that, The preprocessing operation of the real-time running data specifically includes: Perform data cleaning to remove redundant information and erroneous records from the original data; Missing value imputation is performed. For data loss caused by sensor failure or transmission interruption, time series interpolation methods are used to supplement it. The time series interpolation methods include linear interpolation, Lagrange interpolation, or cubic spline interpolation. Perform outlier detection, identifying and correcting data points that significantly deviate from the normal range using statistical or machine learning methods. The statistical methods include the three-standard-deviation method and box plot method; the machine learning methods include the isolated forest algorithm and the local outlier algorithm. Perform timestamp alignment to ensure that data from different sensors and actuators are accurately synchronized over time. The preprocessing operation also includes data standardization, which scales the preprocessed data to a preset numerical range or makes its mean zero and variance one, in order to eliminate the influence of different units on subsequent model training and analysis. The data standardization adopts the minimum-maximum normalization method or the Z-score standardization method.

4. The method according to claim 1, characterized in that, The construction of the multi-process coupled digital twin model of the prefabricated wastewater treatment plant specifically includes: A mechanism model module is constructed, which includes biochemical reaction kinetics models, hydraulic models, and material-energy balance models for each treatment unit of the prefabricated wastewater treatment plant. The model parameters are calibrated offline using historical operating data and corrected in real time using an online adaptive algorithm. The mechanism model module includes enhanced anaerobic-aerobic process model, biological contact oxidation process model, aerobic granular sludge process model and photosynthetic bacteria treatment process model. Each process model is customized based on the activated sludge model framework, including key components such as soluble matrix, granular matrix, dissolved oxygen, nitrogen compounds, phosphorus compounds, activated sludge and inert sludge. A data-driven model module is constructed, which uses historical operational data to train a deep learning model to capture nonlinear and time-dependent features that are difficult for mechanistic models to describe. In this data-driven model module, the deep learning model employs a long short-term memory network, a gated recurrent unit, or a transformer model based on an attention mechanism. The input to the deep learning model includes preprocessed sensor data, actuator state data, and external disturbance data. A multi-process coupled hybrid model is constructed, which integrates the mechanism model module and the data-driven model module, and clarifies the material flow, energy flow and hydraulic connection relationships between different process modules; The multi-process coupled hybrid model uses a directed graph structure to represent the process flow of the prefabricated wastewater treatment plant. The nodes of the directed graph represent each treatment unit, and the edges of the directed graph represent the material exchange paths between units. The multi-process coupled hybrid model uses a Kalman filter or an extended Kalman filter to weight and fuse the prediction results of the mechanistic model with the prediction results of the data-driven model, and uses sensor observation data in real time to update the model state and correct errors.

5. The method according to claim 1, characterized in that, The specific implementation of real-time state synchronization, prediction and simulation, and intelligent optimization decision-making using the multi-process coupled digital twin model includes: Real-time state synchronization is performed to accurately map the real-time operation data of the prefabricated wastewater treatment plant to the corresponding state variables of the digital twin model. The real-time state synchronization adopts an unscented Kalman filter or a particle filter to fuse sensor observation data with the prediction data of the multi-process coupled digital twin model. The system performs prediction and simulation, based on the current synchronized digital twin model state, to simulate the operational trends of the prefabricated wastewater treatment plant under different influent loads, environmental conditions, and control strategies over a future period. This prediction and simulation includes multi-step predictions of key indicators such as effluent quality, energy consumption, carbon source consumption, and sludge production. The system executes intelligent optimization decisions, generating optimal operation control strategies based on prediction and simulation results, combined with preset optimization objectives and constraints. The intelligent optimization decisions employ model predictive control algorithms and combine them with genetic algorithms or particle swarm optimization algorithms to iteratively solve the future control variable sequence.

6. The method according to claim 1, characterized in that, The fault diagnosis and early warning based on the deviation between the real-time operating data and the digital twin model prediction data specifically includes: Continuously monitor the deviation between the real-time operation data of the prefabricated wastewater treatment plant and the predicted data of the multi-process coupled digital twin model; When the deviation exceeds a preset threshold, an anomaly event detection is triggered. The anomaly event detection adopts a statistical process control method or a deep learning anomaly detection algorithm. Root cause analysis is performed on detected abnormal events, employing knowledge graph-based reasoning or causal inference methods to identify the source and type of the fault; and Based on the fault diagnosis results, warning information is issued to operators through visual interfaces, audible and visual alarms, SMS or email, and fault handling suggestions are provided.

7. A digital twin monitoring system for multi-process coupled operation of a prefabricated wastewater treatment plant, characterized in that, include: The data acquisition and preprocessing module is used to acquire multi-source heterogeneous operational data from physical wastewater treatment plants and perform cleaning, conversion and standardization processing. The digital twin modeling module is used to build and calibrate hybrid digital twin models with multiple coupled processes online. The simulation and prediction module is used for simulation analysis of operational status and prediction of future trends based on digital twin models. The intelligent monitoring and diagnostic module is used to monitor the operation of the wastewater treatment plant in real time, identify anomalies, and perform fault diagnosis and performance evaluation. The optimization decision-making module is used to generate optimized operating strategies and parameter recommendations based on monitoring, diagnosis, and prediction results. The control command output module is used to convert optimization strategies into executable control commands and send them to the physical wastewater treatment plant; as well as The human-computer interaction interface module is used to provide a platform for user operation and information display.

8. The system according to claim 7, characterized in that, The data acquisition and preprocessing module includes a data interface unit, a data cleaning unit, and a feature engineering unit. The data interface unit interacts with the online sensors, industrial automation control system, and laboratory information management system of the physical wastewater treatment plant through an industrial communication interface. After receiving the raw data, the data cleaning unit performs outlier detection, missing value filling, data denoising, and timestamp alignment. as well as The feature engineering unit performs temporal feature extraction and data standardization or normalization on the cleaned data.

9. The system according to claim 8, characterized in that, The digital twin modeling module includes a mechanism model library, a data-driven model library, a model integration unit, a parameter calibration unit, and a data assimilation unit. The mechanism model library stores mathematical mechanism models for each process unit in the prefabricated wastewater treatment plant; The data-driven model library stores pre-trained machine learning and deep learning models; The model integration unit couples the heterogeneous models to construct a unified multi-process coupled digital twin model; The parameter calibration unit uses historical operating data and an optimization algorithm to calibrate key parameters in the mechanism model; as well as The data assimilation unit uses the Kalman filter algorithm to correct the state variables and parameters of the digital twin model online using real-time acquired data.

10. The system according to claim 8, characterized in that, The intelligent monitoring and diagnosis module includes an anomaly detection unit, a fault diagnosis engine, and a performance evaluation unit. The anomaly detection unit continuously monitors the deviation between the real-time operating data of the physical wastewater treatment plant and the predicted values ​​of the digital twin model, and uses anomaly detection algorithms to identify operational anomalies. The fault diagnosis engine combines a pre-set fault knowledge graph, expert rules, and causal reasoning mechanism to conduct in-depth analysis of detected anomalies, locate the fault source, and provide root cause analysis. as well as The performance evaluation unit is based on a multi-index evaluation system to quantitatively evaluate the operating efficiency, environmental benefits, economic benefits, and stability of the wastewater treatment plant.

Images