An accurate aeration system based on ASM model and AI

Abstract

This application relates to the field of wastewater treatment technology and discloses a precise aeration system based on the ASM model and AI. It includes front-end water quality sensing equipment, a PLC control system, a host computer monitoring system, an intelligent aeration server, and a blower unit. The front-end equipment collects influent data and environmental parameters; the PLC receives data and executes instructions; the server runs the ASM2D mechanistic model and LSTM algorithm, executing a two-stage coupled operation logic. In the first stage, the ASM2D mechanistic model is used in conjunction with engineering formulas for actual control and data accumulation. In the second stage, intermediate parameters are output from the ASM2D mechanistic model and used as input to predict the aeration rate using the LSTM algorithm. This invention, through the coupling of a mechanistic model and deep learning algorithm, achieves prediction and closed-loop control of aeration rate under complex operating conditions, helping to solve the control lag problem under nonlinear time sequences, ensuring effluent water quality meets standards while reducing operating energy consumption.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, specifically to a precision aeration system based on the ASM model and AI. Background Technology

[0002] Wastewater treatment is a high-energy-consuming industrial process. Aerobic aeration is a core component of biological wastewater treatment, and the electricity consumption of the aeration system accounts for a significant proportion of the total energy consumption of a wastewater treatment plant. Precise control of the aeration rate directly affects the effluent quality and the operating costs of the wastewater treatment plant. Insufficient aeration will lead to substandard effluent quality, while excessive aeration not only wastes electricity but also causes sludge aging and difficulty in flocculation, and reduces denitrification efficiency after mixed liquor recirculation. Therefore, achieving precise control of the aeration system is a key requirement for wastewater treatment plants to reduce operating costs and improve treatment efficiency.

[0003] Existing wastewater treatment plant aeration control mainly relies on human experience or fixed design parameters. The activated sludge model ASM2D is widely used to simulate wastewater treatment processes, and commercial software based on the activated sludge model is also used in process parameter design. In addition, artificial intelligence algorithms such as neural networks and long short-term memory networks are also used to simulate and predict aeration process parameters in the aerobic zone. Artificial intelligence algorithms attempt to approximate the expected values ​​by mathematically solving optimization problems, and try to solve the nonlinear control problem in the wastewater treatment process.

[0004] However, due to the significant fluctuations in the influent quality and quantity of wastewater treatment plants, relying solely on human experience or fixed parameters cannot adapt to changes in the actual operating environment in real time. Furthermore, activated sludge models cannot directly output the specific air supply required to maintain the target dissolved oxygen level in the aerobic tank; they can only provide a setpoint for dissolved oxygen concentration. Moreover, integrating existing commercial software into online control systems involves high interface development costs and periodic parameter calibration costs, making real-time closed-loop control difficult. Artificial intelligence algorithms require high-quality, high-quantity, and high-computing-power historical data, while on-site online process water quality sensors at wastewater treatment plants are susceptible to environmental pollution and drift, resulting in poor long-term stability and low accuracy. This leads to a lack of high-quality, continuous training data for artificial intelligence algorithms. Finally, purely data-driven models lack constraints at the physical mechanism level, making it difficult for them to maintain stable predictive accuracy when facing large fluctuations in influent load or sudden water quality shocks, leading to deviations in control strategies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a precision aeration system based on the ASM model and AI. It aims to solve the problems in existing technologies where a single mechanism model cannot directly output aeration control quantities, and a single data-driven model heavily relies on the quality of on-site sensor data and lacks mechanistic constraints, resulting in low aeration control accuracy and poor system stability under fluctuating influent load conditions.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention provides a precision aeration system based on the ASM model and AI, comprising: Front-end water quality sensing equipment, PLC control system, sewage treatment plant upper computer monitoring system, intelligent aeration system server, and related equipment for blower units.

[0008] The front-end water quality sensing equipment is configured to collect influent flow rate data, influent water quality data, and environmental parameters during the wastewater treatment process. The influent water quality data includes chemical oxygen demand (COD) and ammonia nitrogen concentration. The influent flow rate data consists of continuous instantaneous flow rate provided by a flow meter installed in the main influent pipe or open channel. The environmental parameters include dissolved oxygen concentration, temperature, pH value, and pressure data in the aerobic tank.

[0009] The PLC control system is configured to receive the collected data and execute control commands. Its hardware mainly consists of a central processing unit, analog input modules, analog output modules, and a communication module. The blower unit includes a blower unit equipped with a variable frequency speed controller, an electric regulating valve, and an air flow meter. These components are connected to the PLC control system to achieve adjustment and closed-loop control of the aeration volume. The wastewater treatment plant's supervisory control and data acquisition (SCADA) system is configured to transmit data between the PLC control system and the intelligent aeration system server.

[0010] The intelligent aeration system server runs an ASM2D mechanism model and an LSTM algorithm, and realizes the prediction and control of aeration volume through the coupling logic of the mechanism model and the data-driven algorithm.

[0011] In terms of mechanism simulation, the server constructs a biochemical reaction simulation model based on the ASM2D mechanism model. This model divides the wastewater treatment biochemical unit into multiple completely mixed reaction tanks and establishes biochemical reaction kinetic equations based on the mass balance principle. The mass balance principle states that the rate of change of component mass over time in any reaction tank is equal to the sum of the component mass rate flowing into this reaction tank from the previous reaction tank, the component mass rate brought into this reaction tank by sludge return and mixed liquor return, minus the component mass rate flowing out of this reaction tank, and plus the component mass rate generated by the biochemical reaction within this reaction tank.

[0012] Based on the aforementioned kinetic equations, the system calculates the theoretical total oxygen demand required to maintain process stability using the dissolved oxygen balance principle. The dissolved oxygen balance principle states that the rate of change of dissolved oxygen concentration in the reaction tank is equal to the difference between the dissolved oxygen flux brought in by the upstream water and the dissolved oxygen flux flowing out of the reaction tank, minus the oxygen consumption rate, and plus the oxygen transfer rate. During the calculation, the rate of change of dissolved oxygen concentration is set to zero, and the theoretical total oxygen demand required to maintain stable dissolved oxygen in the aerobic tank is calculated through integration. Combined with physical parameters such as oxygen transfer efficiency, water quality, temperature, pressure, and the immersion depth of the aeration heads, the theoretical air supply of the blower is calculated.

[0013] In terms of intelligent prediction, this invention utilizes the LSTM algorithm to construct an aeration volume prediction model. The algorithm takes the influent flow rate, aerobic tank water quality parameters including chemical oxygen demand (COD) and nitrogen and phosphorus indices, sludge concentration, environmental parameters, dissolved oxygen at the previous moment, and equipment status parameters as input variables, and the predicted dissolved oxygen value in the aerobic zone as the output variable.

[0014] The coupled execution logic of the execution mechanism model and data-driven algorithm in this invention is specifically divided into a data accumulation and model training stage, and an online intelligent control stage: During the data accumulation and model training phase, the system uses the ASM2D mechanism model combined with the engineering aeration calculation formula to calculate the theoretical aeration volume, and uses the theoretical aeration volume as the execution command to actually control the blower unit. In this process, the system synchronously records the operation records containing process water quality indicators and equipment execution data as the training dataset, performs supervised training on the LSTM algorithm, and establishes the mapping relationship between input features and aeration control quantities. During the online intelligent control phase, the ASM2D mechanism model runs online and performs a self-check on its simulation accuracy. After passing the self-check, it generates intermediate state parameters that characterize the microscopic state of the biochemical system. The LSTM algorithm takes the real-time monitoring data and the intermediate state parameters as joint inputs and, based on the mapping relationship learned during the training phase, directly generates aeration control commands to drive the blower unit.

[0015] In addition, the present invention is configured with model parameter self-calibration and self-learning logic to adapt to changes in actual working conditions.

[0016] The system tracks the quality of the effluent under model control in real time. When the absolute error between the simulated effluent quality and the measured data is greater than 10%, parameter self-calibration is triggered.

[0017] The calibration process employs a hierarchical optimization strategy: Prioritize the adjustment of high-sensitivity parameters: Based on the parameter sensitivity ranking table (e.g., the sensitivity of the maximum specific growth rate of nitrifying bacteria to ammonia nitrogen is >1.85), prioritize the calibration of high-impact parameters such as the maximum specific growth rate of nitrifying bacteria and the denitrification rate. Incremental step size control: The single parameter adjustment range is ≤5%, and the model accuracy is verified after each adjustment until the simulation error is consistently <5%.

[0018] Meanwhile, the system continuously records dissolved oxygen, aeration air volume, simulated water quality and environmental parameters during operation, stores the data after unifying them to the same time series, and periodically uses the accumulated data to train and update the LSTM model to maintain the accuracy of the prediction algorithm.

[0019] During the control execution phase, the intelligent aeration system server writes back the calculated aeration volume data to the sewage treatment plant's upper computer monitoring system, which then converts it into control commands and sends them to the PLC control system. This, in turn, adjusts the blower frequency and valve opening to achieve intelligent and precise aeration in the sewage treatment process.

[0020] This invention provides a precise aeration system based on the ASM model and AI. It offers the following advantages: 1. This invention discloses a precision aeration system based on the ASM model and AI. By coupling a mechanistic model with a data-driven algorithm, it overcomes the limitations of single-model control. The ASM2D mechanistic model provides target dissolved oxygen values ​​and process parameters that conform to the process principles based on biochemical reaction kinetics, ensuring that the effluent quality meets the set requirements. Simultaneously, it utilizes the LSTM algorithm's ability to process nonlinear time-series data to establish a mapping relationship between aeration rate and dissolved oxygen. This coupling method leverages the interpretability of the mechanistic model to constrain the prediction boundary of the AI ​​algorithm, improving the reliability of the aeration control strategy under complex operating conditions such as large fluctuations in influent water quality.

[0021] 2. This invention employs a reverse optimization logic based on LSTM iterative calculation to determine the aeration rate, solving the lag problem inherent in traditional feedback control in biochemical reactions. The system uses the theoretical air supply calculated by ASM2D as an initial basis, and employs the LSTM algorithm for rapid iteration to deduce the specific aeration air volume required to meet the target dissolved oxygen level. This control mode, combining feedforward prediction and reverse solving, can respond proactively to changes in influent load, reducing dissolved oxygen concentration fluctuations caused by response delays and helping to maintain the stability of the microbial environment within the biochemical tank.

[0022] 3. This invention, by configuring self-calibration of mechanistic parameters and self-learning of AI algorithms, adapts to long-term changes in the operating environment of wastewater treatment plants. When the deviation between actual operating data and model simulation data continues to exceed the threshold, the system automatically corrects the stoichiometric coefficients and kinetic parameters based on sensitivity analysis, and periodically updates the LSTM model using accumulated historical operating data. This helps to reduce the impact of changes in activated sludge properties or seasonal environmental fluctuations on the accuracy of the control system, reduces the frequency of manual model maintenance, and extends the period of stable system operation. Attached Figure Description

[0023] Figure 1 This is a schematic diagram illustrating the construction of the LSTM prediction aeration rate algorithm according to an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the coupling process between the ASM2D mechanism model and the LSTM algorithm-based aeration rate prediction model in an embodiment of the present invention. Figure 3 This is a comparison chart of the simulated effluent values ​​of various indicators in this embodiment of the invention with the measured values ​​of the wastewater treatment plant and the simulated values ​​of WEST software; The top figure shows a comparison of simulated and measured changes in COD, and the bottom figure shows a comparison of simulated and measured changes in NH4-N. Figure 4 This is a schematic diagram of the training and testing results of the LSTM algorithm in an embodiment of the present invention; Figure 5 This is a flowchart illustrating the model parameter self-calibration and algorithm self-learning process in an embodiment of the present invention. Figure 6 This is a schematic diagram of the intelligent aeration control system architecture for a wastewater treatment plant, as described in an embodiment of the present invention. Figure 7 This is a comparison chart of the target control value and the actual control value of dissolved oxygen in an embodiment of the present invention. Detailed Implementation

[0024] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Examples 1-3: Example 1: See attached document Figure 6 This invention provides a precision aeration system based on the ASM model and AI, comprising: Front-end water quality sensing equipment, PLC control system, sewage treatment plant upper computer monitoring system, intelligent aeration system server, and related equipment for blower units.

[0026] The front-end water quality sensor collects real-time data during the wastewater treatment process. This data includes influent flow rate, influent water quality data (e.g., COD, ammonia nitrogen), and environmental parameters (e.g., water temperature, pH, aeration pressure). The front-end water quality sensor communicates with the PLC control system.

[0027] The PLC control system serves as the core for on-site control and data acquisition. It connects to front-end water quality sensors via its input terminals to collect and upload information such as influent flow rate and water quality data; and it connects to relevant equipment in the blower unit via its output terminals to execute control commands. The PLC control system communicates with the wastewater treatment plant's supervisory control and data acquisition (SCADA) system.

[0028] The wastewater treatment plant's supervisory control and data acquisition (SCADA) system functions as a monitoring and interaction platform. It communicates with both the PLC control system and the intelligent aeration system server. This system receives data uploaded by the PLC control system and transmits it to the intelligent aeration system server; simultaneously, it receives data written back from the intelligent aeration system server and sends control commands to the PLC control system.

[0029] The intelligent aeration system server serves as the core computing platform. Internally, it runs an ASM2D mechanistic model and an LSTM algorithm. This server combines data such as carbon source addition and the current operating status of the wastewater treatment system. Using the ASM2D mechanistic model, it simulates and outputs the target control values ​​for dissolved oxygen in the aerobic zone and process water quality parameters. Then, using the LSTM algorithm, it predicts the corresponding aeration rate by receiving the data output from the mechanistic model.

[0030] The blower unit includes variable frequency fans, fixed frequency fans, air flow meters, and aeration zone valves. These devices operate according to instructions from the PLC control system, achieving precise air distribution by finely adjusting the fan operating frequency and the opening degree of the aeration zone valves.

[0031] The specific workflow of this system is as follows: First, the water volume and water quality data are collected by the front-end water quality sensing equipment and transmitted to the PLC control system.

[0032] Next, the PLC control system uploads the collected data to the sewage treatment plant's upper-level monitoring system.

[0033] Then, the intelligent aeration system server obtains data from the sewage treatment plant's supervisory control system and uses the ASM2D mechanism model to simulate the target control value of dissolved oxygen and process water quality parameters required in the aerobic zone.

[0034] Subsequently, the LSTM algorithm receives the data output from the mechanistic model and predicts the corresponding aeration rate. This prediction process involves repeated simulations until the predicted dissolved oxygen value and the target dissolved oxygen control value are within a certain control accuracy range, thereby determining the aeration rate in reverse.

[0035] Finally, the intelligent aeration system server writes the determined aeration volume data back to the sewage treatment plant's upper-level monitoring system, which then sends control commands to the PLC control system to complete the intelligent aeration control.

[0036] See attached document Figure 6 The front-end water quality sensing equipment is deployed at key process nodes such as the inlet and biological treatment tank of the sewage treatment plant. As the sensing layer of the system, it is responsible for converting the water quality, water quantity and environmental conditions of the physical world into electrical signals and transmitting them to the PLC control system.

[0037] The front-end water quality sensing equipment specifically includes water quality monitoring instruments, water quantity monitoring instruments, and environmental parameter sensors.

[0038] Water quality monitoring instruments are installed at the inlet channel of the sewage treatment plant, after the coarse screen or before the fine screen, to obtain inlet water load parameters.

[0039] Specifically, the water quality monitoring instrument includes a sampling unit, a pretreatment unit, and an analysis unit. The sampling unit draws water samples from the center point of the main flow channel using a submersible pump or a self-priming pump. The water samples are then filtered by the pretreatment unit to remove coarse suspended solids before entering the analysis unit.

[0040] In this embodiment, the water quality monitoring instruments specifically include an online COD analyzer and an online ammonia nitrogen analyzer. The online COD analyzer is used to detect chemical oxygen demand and converts the measured COD concentration value into a 4-20mA analog current signal or digital signal, which serves as the input for the ASM2D mechanistic model to calculate the influent carbon source load. The online ammonia nitrogen analyzer is used to detect ammonia nitrogen concentration, and this ammonia nitrogen concentration data serves as a key initial condition for simulating the nitrification process in the ASM2D mechanistic model.

[0041] Water flow monitoring instruments are installed on the main inlet pipe or open inlet channel to provide continuous instantaneous flow data. Depending on the type of inlet pipeline, electromagnetic flow meters or ultrasonic open channel flow meters can be used. For scenarios where the inlet is a full-pipe flow, an electromagnetic flow meter is used. This flow meter is installed on the outlet pipe of the inlet booster pump station and calculates the flow velocity and flow rate by measuring the induced electromotive force generated when conductive wastewater cuts magnetic lines of force. For scenarios where the inlet is a gravity-flow open channel, an ultrasonic open channel flow meter is used. This ultrasonic flow meter is used in conjunction with a Parshall flume or triangular weir, with its probe installed above the flume or weir. It measures the liquid level height non-contactly and converts it into flow data. The inlet flow data is a fundamental parameter for the PLC control system to calculate the total gas demand and for the LSTM algorithm model to divide the time series features.

[0042] Environmental parameter sensors are directly submerged in the aerobic zone of the biological treatment tank or installed on the air supply pipeline to monitor the reaction environment and equipment status. Specifically, these sensors include a dissolved oxygen sensor, a temperature sensor, a pH meter, and a pressure sensor. The dissolved oxygen sensor is fixedly installed on a bracket in the middle or end of the aerobic tank, with its probe submerged below the liquid surface. This sensor has an automatic cleaning function, using scraping or air blowing to prevent biofilm adhesion from affecting measurement accuracy. Its real-time dissolved oxygen concentration output is the core controlled variable for closed-loop feedback control of this system. The temperature sensor and pH meter are typically integrated or installed independently in the aerobic tank. Temperature data is used to correct the reaction kinetic coefficients in the ASM2D mechanistic model, and pH values ​​are used to monitor the acid-base balance of the biological treatment system. The pressure sensor is installed on the blower's main outlet pipe or aeration branch pipe to monitor the back pressure of the air duct network. This pressure parameter is used by the PLC control system to determine whether the aeration discs are blocked and to assist in correcting the blower's operating frequency.

[0043] All the aforementioned front-end water quality sensing devices are equipped with signal transmitters, which are connected to the analog input module or communication module of the PLC control system via shielded twisted-pair cables. For parameters such as instantaneous flow rate, dissolved oxygen concentration, and pressure that change rapidly, a 4-20mA hardwired connection is preferred to ensure real-time data transmission. For devices with longer data update cycles, the Modbus-RTU fieldbus protocol is used to communicate with the PLC control system.

[0044] In the above specific implementation, by scientifically deploying water quality monitoring instruments, water quantity monitoring instruments, and environmental parameter sensors at the inlet and biological treatment tank, and by adopting differentiated signal transmission strategies, real-time and accurate acquisition and signal conversion of influent components, hydraulic load, and reaction environment parameters of the wastewater treatment system are achieved. This helps to eliminate data blind spots and lags in traditional control modes, and provides continuous and multi-dimensional input data support for subsequent ASM2D mechanism model simulation and LSTM algorithm prediction in the intelligent aeration system server, thereby ensuring the accuracy and real-time performance of the intelligent aeration control strategy.

[0045] See attached document Figure 6 The PLC control system is deployed in the field electrical control cabinet. As the execution hub connecting physical equipment and upper management system, its input end is connected to the front-end water quality sensing equipment, its output end is connected to the relevant equipment of the blower unit, and it is connected to the upper computer monitoring system of the sewage treatment plant through the industrial Ethernet communication interface.

[0046] The PLC control system adopts a modular design in terms of hardware, mainly consisting of a central processing unit, analog input module, analog output module and communication module.

[0047] The analog input module receives current or voltage signals from front-end water quality sensors, specifically covering key parameters such as influent flow rate, chemical oxygen demand (COD), ammonia nitrogen concentration, and dissolved oxygen concentration. The central processing unit (CPU) performs analog-to-digital conversion, filtering, noise reduction, and range linearization on the received raw signals, converting them into standard engineering quantities. Subsequently, the real-time data is uploaded to the wastewater treatment plant's supervisory control and data acquisition (SCADA) system via the communication module, based on a common industrial protocol, for use in subsequent model calculations.

[0048] In terms of control execution, the PLC control system receives optimized control commands from the intelligent aeration system server, forwarded by the wastewater treatment plant's supervisory control and data acquisition (SCADA) system. The central processing unit (CPU) parses these commands into corresponding control signals and sends frequency adjustment signals and valve opening signals to relevant equipment in the blower unit via the analog output module, thereby changing the air supply of the aeration system. Furthermore, the PLC control system integrates network outage protection logic. When a communication heartbeat loss with the wastewater treatment plant's SCADA system is detected, the system automatically locks the last valid control parameters or switches to a preset constant pressure operating mode until communication is restored.

[0049] In the above specific implementation, by configuring a PLC control system with signal preprocessing and network outage protection functions, high-fidelity acquisition of underlying sensing data and accurate implementation of upper-level control commands are achieved, which helps to ensure the stability of the data transmission link and the continuity and safety of the sewage treatment process operation under sudden communication failures in complex industrial environments.

[0050] See attached document Figure 6 The sewage treatment plant's supervisory control and data acquisition (SCADA) system is deployed in the central control room of the sewage treatment plant. As a data interaction platform connecting the lower-level control network and the upper-level computing network, the sewage treatment plant's SCADA system communicates with the PLC control system and the intelligent aeration system server.

[0051] The wastewater treatment plant's supervisory control and data acquisition (SCADA) system comprises an industrial control computer, display terminals, and network switching equipment in its hardware architecture. In terms of software configuration, the system runs industrial configuration monitoring software and a real-time database system. The industrial configuration monitoring software establishes a data connection with the PLC control system via the industrial Ethernet protocol or OPC communication standard. It is configured to collect and display in real-time data on influent flow rate, influent water quality, and environmental parameters uploaded by the PLC control system, and stores this data in a historical database for querying and trend analysis.

[0052] In terms of data interaction logic, the sewage treatment plant's supervisory control and data acquisition (SCADA) system performs the tasks of data transmission and instruction conversion.

[0053] On the one hand, the sewage treatment plant's supervisory control and data acquisition system will perform protocol parsing and standardization on the real-time operating data obtained from the PLC control system, and write it into the designated data receiving area of ​​the intelligent aeration system server through the ODBC (Open Database Connectivity) interface or TCP / IP communication protocol, providing input boundary conditions for the calculation of the ASM2D mechanism model.

[0054] On the other hand, the wastewater treatment plant's supervisory control and data acquisition (SCADA) system reads in real time the aeration volume control data and process parameters, which are determined based on model predictions and written back from the intelligent aeration system server. This data is then mapped to specific PLC address register values ​​and sent to the PLC control system for execution. Furthermore, the SCADA system provides a human-machine interface, allowing operators to switch between automatic and manual control modes and set relevant alarm thresholds.

[0055] In the above specific implementation, by deploying a sewage treatment plant upper computer monitoring system that integrates configuration monitoring and data management functions, bidirectional real-time interaction between the underlying production data and the top-level algorithm model is realized. This helps to eliminate data communication barriers between different control levels, provides a continuous and complete data stream for the intelligent aeration system server, and ensures the accuracy and visualization of control commands.

[0056] See attached document Figure 6 The intelligent aeration system server is deployed in the central control room or a separate server room. The intelligent aeration system server communicates with the sewage treatment plant's upper computer monitoring system through a local area network or fiber optic channel.

[0057] As a specific implementation method, the intelligent aeration system server adopts an industrial-grade rack-mount server or graphics workstation at the hardware level. The server is equipped with a multi-core central processing unit, random access memory, and an independent graphics processing unit. The independent graphics processing unit is configured to perform matrix-parallel operations and model training tasks for deep learning algorithms. In terms of software, the intelligent aeration system server runs a real-time operating system and deploys mathematical simulation programs and a deep learning inference engine written in Python or C++.

[0058] The intelligent aeration system server acquires real-time influent data from the wastewater treatment plant's supervisory control and data acquisition system (SCADA) via an ODBC database interface or TCP / IP communication protocol, according to a preset control cycle, such as every 10 to 30 minutes. This real-time influent data includes chemical oxygen demand (COD) concentration, ammonia nitrogen concentration, influent flow rate, and biological tank water temperature, all collected by front-end water quality sensors. The intelligent aeration system server runs an internally stored ASM2D mechanistic model, mapping the aforementioned real-time influent data to the input boundary conditions of the ASM2D mechanistic model. The ASM2D mechanistic model solves the activated sludge reaction kinetic equations using numerical integration, simulating the organic matter degradation and nitrification / denitrification biochemical reaction processes within the biological tank. It calculates the theoretical oxygen demand required to ensure effluent quality meets standards under current operating conditions and determines the target dissolved oxygen concentration in the aerobic zone of the biological tank accordingly.

[0059] The intelligent aeration system server further utilizes the LSTM (Long Short-Term Memory) network model. The server combines the target dissolved oxygen concentration output from the ASM2D mechanistic model, along with virtual process parameters such as nitrification rate, COD degradation rate, and heterotrophic bacteria growth rate generated by the model simulation, with influent flow rate sequences, historical dissolved oxygen concentration sequences, and blower historical operating frequency sequences stored in the historical database. This constructs a multi-dimensional time-series feature vector, which is then input into the LSTM network model. This design replaces the real-time monitoring function of high-failure-rate sensors such as online ammonia nitrogen analyzers and phosphate analyzers with virtual parameters output from the mechanistic model, forming a soft measurement framework constrained by the ASM mechanism.

[0060] The LSTM (Long Short-Term Memory) network model utilizes the input gate, forget gate, and output gate structure in the LSTM network to extract the nonlinear lag features in time series data and predict the blower operating frequency command and regulating valve opening command to meet the target dissolved oxygen concentration.

[0061] The intelligent aeration system server sends the calculated blower operating frequency command and regulating valve opening command to the wastewater treatment plant's upper-level monitoring system, which then forwards them to the PLC control system for execution. Furthermore, the intelligent aeration system server is equipped with model self-calibration logic. The server compares the actual dissolved oxygen concentration with the target dissolved oxygen concentration predicted by the model in real time. When the absolute error between the two continuously exceeds a preset deviation threshold, the server triggers an online learning program, using the backpropagation algorithm to update the weight matrix of the LSTM (Long Short-Term Memory) network model or to correct the key kinetic parameters of the ASM2D mechanistic model to rectify model bias.

[0062] In the above specific implementation, by constructing a coupled computational architecture of ASM2D mechanism model and LSTM deep learning model in the intelligent aeration system server, and using GPU hardware acceleration to realize real-time simulation and prediction of complex biochemical processes, the transformation from experience-dependent open-loop control to model-predictive closed-loop control is realized. This helps to output aeration control commands in real time when dealing with drastic fluctuations in influent load, and reduce blower energy consumption while ensuring stable effluent water quality.

[0063] See attached document Figure 6 The blower unit and related equipment are installed in the blower room of the sewage treatment plant and in the air duct system extending to the aerobic zone of the biological treatment tank. The blower unit and related equipment serve as the end actuators of this system. The electrical control terminals of the blower unit and related equipment are connected to the output and input modules of the PLC control system. The air output terminals of the blower unit and related equipment are connected to the microporous aerators at the bottom of the biological treatment tank through pipelines.

[0064] The main hardware components of the blower unit include blower units, electric regulating valves, and air flow meters. The blower unit serves as the air supply device, employing either an air-suspended centrifugal blower or a magnetically suspended centrifugal blower. The blower unit integrates a variable frequency speed controller, configured to receive frequency or speed setting signals from the PLC control system. When the PLC control system outputs adjustment signals based on instructions from the intelligent aeration system server, the blower unit adjusts the motor speed to change the total air supply. Simultaneously, the blower unit is configured to monitor its own operating parameters in real time, including motor speed, current, exhaust pressure, and exhaust temperature, and feed these parameters back to the PLC control system.

[0065] Electric regulating valves are installed on branch pipes leading to each aerobic zone of the biological treatment tank. These valves are configured to regulate the airflow entering specific biological reaction zones. The electric regulating valves feature linear regulation characteristics and receive opening control signals from the PLC control system. These signals are typically 4-20mA analog signals or fieldbus commands. The PLC control system independently adjusts the valve opening of the corresponding branch pipe based on the dissolved oxygen deviation in different aerobic zones, thereby achieving on-demand distribution of aeration air. The electric regulating valves are also equipped with position feedback devices, which transmit the actual valve opening signal back to the PLC control system to verify the accuracy of valve operation.

[0066] Air flow meters are installed on the main outlet pipe of the blower unit and the branch pipes of each aerobic zone. These air flow meters are configured to collect actual airflow data within the pipeline. Generally, thermal mass flow meters are used, capable of directly measuring gas mass flow under standard conditions without the need for additional temperature and pressure compensation. The air flow meters transmit the collected real-time flow data to the PLC control system. The PLC control system compares this real-time flow data with the target aeration rate calculated by the intelligent aeration system server. It then uses a PID algorithm to perform closed-loop corrections on the blower unit's frequency and the opening of the electric regulating valve until the actual airflow data matches the target aeration rate.

[0067] In the above specific embodiments, by configuring the blower unit related equipment including a high-efficiency variable frequency blower, a linear electric regulating valve and a high-precision air flow meter, an execution architecture of main pipe pressure stabilization and branch pipe flow regulation and a flow closed-loop control architecture are constructed. This realizes the physical-level precise execution of the intelligent aeration system server optimization instructions, which helps to ensure that the dissolved oxygen concentration in each area of ​​the biological tank can respond quickly and accurately to changes in the influent load, and minimizes the energy waste caused by over-aeration.

[0068] Example 2: See attached document Figure 1 Appendix Figure 2 and appendix Figure 5 The present invention further provides the logic for constructing and running an aeration volume prediction model within the intelligent aeration system server, the logic specifically including: The logic for constructing a biochemical reaction simulation model based on the ASM2D mechanism model, the logic for constructing an aeration volume prediction algorithm based on the LSTM algorithm, the coupled operation logic of the mechanism model and the artificial intelligence algorithm, and the logic for model parameter self-calibration and algorithm self-learning.

[0069] The logic for constructing a biochemical reaction simulation model based on the ASM2D mechanism model. The intelligent aeration system server divides the wastewater treatment biochemical unit into... A fully mixed reaction tank, with the intelligent aeration system server numbering the tanks sequentially according to the water inflow order. equal to 1 to The secondary sedimentation tank is numbered as follows: equal Add 1, the intelligent aeration system server sets the solution volume in the reaction tank to be The intelligent aeration system server defines the influent to any reaction tank as consisting of four parts: The first part is the front-end influent, and the first tank is the system influent; the second part is sludge return; the third part is mixed liquor return; the fourth part is other influent, which includes segmented influent or the addition of carbon source agents.

[0070] Based on the mass balance principle that the change in any component in a reaction tank equals the input minus the output plus the reaction amount, the intelligent aeration system server establishes the following biochemical reaction kinetic equation that varies with time: ; In the formula, For the first Pool effluent components concentration; It is a time variable; This refers to the volume of the solution in the reaction tank. This is the effluent flow rate from the previous reaction tank; The effluent components of the previous reaction tank concentration; The sludge return ratio; This refers to the inflow rate of the first pool system; The concentration of components in the secondary sedimentation tank; The reflux ratio of the mixture; For recirculation pool Component concentration.

[0071] In the initial simulation, the intelligent aeration system server used the default values ​​for reaction kinetic parameters and stoichiometric coefficients from the International Water Association (IWA). All time variables in the ASM2D mechanistic model are expressed in days (d), and flow rates in meters (m). 3 / d, component concentration in g / m 3 The intelligent aeration system server decomposes and preprocesses the influent chemical oxygen demand (COD) and total phosphorus (TP) data. The server then converts the preprocessed data into the data format required by the ASM2D mechanistic model for calculation.

[0072] To maintain a stable dissolved oxygen environment, the total oxygen transfer rate in the aerobic tank must be kept consistent with the dissolved oxygen consumption rate within the tank. The intelligent aeration system server calculates the theoretically required total dissolved oxygen amount using the following integral formula: ; In the formula, Dissolved oxygen concentration; It is a time variable; For the first Pool outlet flow rate; For the first Dissolved oxygen concentration in the pool; For the first Dissolved oxygen concentration in the pool; This refers to the oxygen consumption rate; This represents the oxygen transfer rate.

[0073] The intelligent aeration system server comprehensively considers factors such as oxygen transfer in wastewater, water quality, temperature, pressure, and the immersion depth of aeration heads. It further calculates the aeration pressure and the standard oxygen supply of the aerobic tank. The intelligent aeration system server combines the oxygen utilization rate parameters of the aeration system to obtain the theoretical air supply of the blower.

[0074] See attached document Figure 1 The specific logic for constructing the aeration volume prediction algorithm based on the LSTM algorithm includes: The intelligent aeration system server utilizes deep learning to address long-term dependencies in time-series data. First, the server performs correlation analysis on historical wastewater treatment data, then determines the input parameters based on the oxygen mass transfer mechanism in wastewater treatment. These input parameters include influent flow rate, aerobic tank water quality parameters, sludge concentration, environmental parameters, time-series parameters, and equipment status-related parameters.

[0075] The aerobic tank water quality parameters include chemical oxygen demand (COD), ammonia nitrogen, orthophosphate, and nitrate nitrogen; environmental parameters include water temperature, pH, and aeration pressure; the time-series parameter is the dissolved oxygen in the aerobic tank at the previous moment; and equipment status parameters include aerator cleaning and maintenance time and aeration air volume. Although intelligent aeration requires the final output of aeration air volume data, the dissolved oxygen in the aerobic zone varies over time. Therefore, the intelligent aeration system server selects the predicted dissolved oxygen value in the aerobic zone as the output of the LSTM (Long Short-Term Memory) network algorithm. The intelligent aeration system server repeatedly simulates the process using the LSTM algorithm, using the theoretical air supply calculated by the mechanistic model as the initial value. It iteratively adjusts the input aeration air volume value until the output predicted dissolved oxygen value and the target value are within a certain control precision range, thus determining the input aeration volume in reverse.

[0076] See attached document Figure 1 The specific logic for constructing the aeration volume prediction algorithm based on the LSTM algorithm includes: The intelligent aeration system server utilizes deep learning to address long-term dependencies in time-series data. First, it performs correlation analysis on historical wastewater treatment data, then determines input parameters based on the oxygen mass transfer mechanism in wastewater treatment. These input parameters include macroscopic data monitored by physical sensors and microscopic intermediate parameters calculated by the mechanistic model.

[0077] The macroscopic data monitored by physical sensors include influent flow rate, aerobic tank water quality parameters, sludge concentration, and environmental parameters. Aerobic tank water quality parameters include chemical oxygen demand (COD), ammonia nitrogen, orthophosphate, and nitrate nitrogen; environmental parameters include water temperature, pH, and aeration pressure. Microscopic intermediate parameters are process variables characterizing biochemical reaction rates, calculated online by the ASM2D mechanistic model, including nitrification rate, denitrification rate, and heterotrophic bacteria specific growth rate.

[0078] Regarding the output selection, the intelligent aeration system server no longer outputs the intermediate variable dissolved oxygen setpoint. Instead, it selects the target aeration airflow of the blower directly driving the equipment as the output of the LSTM (Long Short-Term Memory) network algorithm. Through the LSTM algorithm, the intelligent aeration system server uses macroscopic physical data and microscopic intermediate parameters as joint input features. Based on the trained mapping relationship, it directly infers and outputs an aeration airflow control command that meets the current operating conditions.

[0079] See attached document Figure 2 The coupled operational logic of the mechanism model and the artificial intelligence algorithm: This system adopts a two-stage evolutionary coupled control strategy. This strategy includes a data accumulation and model training stage (the first stage) and an online intelligent control stage (the second stage).

[0080] In the first stage, the intelligent aeration system server runs the ASM2D mechanistic model. Based on real-time influent data, the ASM2D model first calculates the biochemical oxygen demand (BOD) under the current operating conditions according to biochemical reaction kinetics. Then, the server calls the built-in traditional engineering aeration calculation formula, combining oxygen transfer efficiency, aeration head submersion depth, atmospheric pressure, and water temperature physical parameters to directly convert the BOD into the theoretical air volume required to maintain process stability. The system directly sends this theoretical air volume as a control command to the PLC control system, driving the on-site blowers to perform actual aeration actions. This mechanism-based control method ensures the safe operation of the wastewater treatment plant before the intelligent model matures. During execution, the system synchronously records full-dimensional time-series data to construct a training dataset. Each sample in the dataset includes real-time influent water quality, influent flow rate, environmental parameters as input features, intermediate parameters calculated by the ASM2D mechanistic model, and the executed air volume, verified in the first stage and capable of maintaining effluent standards, as the output label.

[0081] In the second stage, after the first stage has accumulated data covering typical operating conditions and completed supervised training of the LSTM algorithm, the system switches to online intelligent control mode. At this point, the ASM2D mechanistic model operates in parallel with the online monitor and the actual process, simulating the state of the biochemical tank in real time and performing a self-check on the simulation accuracy. After passing the self-check, the ASM2D mechanistic model outputs intermediate state parameters characterizing the microscopic state of the biochemical system. The LSTM algorithm receives macroscopic data monitored by physical sensors and the microscopic intermediate state parameters output by the ASM2D mechanistic model as a joint input vector. Based on the mapping relationship learned in the first stage, the LSTM algorithm directly predicts and outputs the optimal aeration airflow command for the current operating conditions. This command is directly sent to the PLC control system to drive the equipment, without the need for online iterative calculations or secondary adjustments.

[0082] See attached document Figure 5Model parameter self-calibration and algorithm self-learning logic: The intelligent aeration system server is equipped with a self-optimization program to ensure the long-term operational accuracy of the system. The system tracks and compares the simulated output values ​​of the ASM2D mechanistic model with the actual monitoring values ​​from the front-end water quality sensors in real time. The self-calibration logic is triggered when the absolute error between the model's simulated effluent quality and the measured data consistently exceeds 10%, at which point the system determines the model is inaccurate and initiates the parameter self-calibration program. While the system continuously monitors key indicators such as sludge age change rate and oxygen transfer efficiency as a basis for error attribution analysis, it does not directly use these as the trigger condition for calibration.

[0083] The calibration process employs a hierarchical optimization strategy. First, a primary calibration is performed, prioritizing the adjustment of physical parameters such as the total oxygen transfer coefficient and the temperature correction coefficient. If the error does not converge, a secondary calibration is performed, where the system fine-tunes the high-sensitivity kinetic parameters of the maximum specific growth rate and denitrification rate of nitrifying bacteria based on a parameter sensitivity ranking table. The adjustment process uses incremental step control, with each parameter adjustment increment less than or equal to 5%. Model accuracy is verified after each adjustment step until the simulation error is consistently less than 5%, preventing model distortion caused by parameter drift.

[0084] In terms of AI algorithm self-learning, the intelligent aeration system server periodically uses high-quality data generated during the second phase of operation to incrementally update the LSTM model to adapt to environmental drift caused by equipment aging or seasonal changes, and maintain the accuracy of the prediction algorithm.

[0085] Example 3: See attached document Figure 3 Appendix Figure 4 Appendix Figure 6 and appendix Figure 7The intelligent aeration control system for wastewater treatment constructed in this embodiment of the invention consists of front-end water quality sensing equipment, a PLC control system, a wastewater treatment plant upper-level monitoring system, an intelligent aeration system server, and related equipment for the blower unit. The intelligent aeration system server establishes a communication connection with the wastewater treatment plant upper-level monitoring system and the PLC control system via an industrial Ethernet network. During operation, the PLC control system is responsible for collecting real-time influent flow and water quality data from the biological treatment tank and uploading the data to the intelligent aeration system server. The intelligent aeration system server, combining synchronously acquired carbon source addition data and current wastewater treatment operating status data, uses the ASM2D mechanism model constructed in Embodiment 2 to simulate and calculate the required dissolved oxygen setpoint and corresponding process water quality parameters for the current aerobic zone. The intelligent aeration system server uses the simulated time-series data as an input vector and predicts the blower aeration rate required to maintain the dissolved oxygen setpoint using a trained LSTM algorithm. The intelligent aeration system server writes back the predicted aeration volume control command to the sewage treatment plant's upper computer monitoring system. The sewage treatment plant's upper computer monitoring system then sends the final control command to the PLC control system. The PLC control system adjusts the frequency of the frequency converter or the valve opening in the blower unit according to the command, thereby realizing intelligent aeration closed-loop control.

[0086] See attached document Figure 3 To ensure the reliability of the aforementioned control logic, the intelligent aeration system server needs to verify the accuracy of its integrated ASM2D mechanism model and LSTM algorithm before being officially put into field control. The intelligent aeration system server selects the monthly average influent and effluent data of a wastewater treatment plant as the input boundary conditions for the mechanism model, runs the ASM2D mechanism model for simulation calculations, and compares the simulated effluent quality data with actual detection values ​​and simulation values ​​from commercial software. In the specific verification process, the intelligent aeration system server uses the average influent and effluent data of a wastewater treatment plant in Xi'an City for February as a benchmark. This wastewater treatment plant in Xi'an City adopts the traditional AAO process, with a single unit treatment capacity of approximately 57,000 m³. 3 / d. Calculations show that the monthly average absolute error of the chemical oxygen demand (COD) value of the wastewater treatment plant effluent simulated by the intelligent aeration system server is 6.45 mg / L, lower than the error value of 7.76 mg / L calculated by the WEST commercial simulation software; the monthly average absolute error of the ammonia nitrogen value simulated by the intelligent aeration system server is 0.15 mg / L, lower than the error value of 0.24 mg / L calculated by the WEST commercial simulation software. These verifications demonstrate that the biochemical reaction simulation environment constructed within the intelligent aeration system server, under the same conditions, has an accuracy comparable to internationally renowned software in simulating effluent, providing a foundation for subsequent training data generation.

[0087] See attached document Figure 4After verifying the mechanistic model, the intelligent aeration system server further validated the predictive ability of the LSTM algorithm. The server selected approximately 11,000 historical operational data points from a wastewater treatment plant in Guizhou, deleting blank, missing, or abnormal data or supplementing them using interpolation. The server applied the Savitzky-Golay filtering algorithm to reduce data noise interference with the prediction model and used the minimax normalization method to standardize the data, eliminating differences between data of different dimensions. The server randomly selected 80% of the data as the training set and 20% as the test set, setting the LSTM algorithm structure to a two-layer 128-64 structure, with a dropout rate of 0.5, a learning rate of 0.0001, a time step of 48, 50 iterations, and a batch size of 64. The server compared the training effects of two different data sources. When trained directly using historical data from the actual operation of a wastewater treatment plant in Guizhou, the algorithm's mean squared error (MSE) was 1.0420 mg / L, mean absolute error (MAO) was 0.7781 mg / L, and coefficient of determination (COD) was 0.7049. The percentage of data where the predicted value deviated from the actual value by less than 0.3 mg / L was 78.64%. However, when trained using augmented historical data generated by the ASM2D mechanistic model, the algorithm's MSE decreased to 0.1274 mg / L, the MAO decreased to 0.2465 mg / L, the COD increased to 0.9561, and the percentage of predicted results deviating from the actual results by less than 0.3 mg / L increased to 96.77%. This demonstrates that dual-drive training using data generated by the mechanistic model helps improve the weak regularity problem inherent in relying solely on actual historical data.

[0088] See attached document Figure 7 Based on the above verification results, the intelligent aeration system server underwent a field industrial application test at a wastewater treatment plant in Guizhou Province. This wastewater treatment plant in Guizhou treats approximately 9000 m³ of domestic sewage per day. 3 / d, using oxidation ditch process. The intelligent aeration system server is set with a predictive control cycle of 15 minutes and a control dead zone of 0.3 mg / L. That is, when the error between the measured value and the set value of dissolved oxygen in the aerobic tank does not exceed 0.3 mg / L, the intelligent aeration system server maintains the blower output constant to avoid frequent equipment adjustments. In terms of dissolved oxygen control stability, compared with the fluctuation of dissolved oxygen between 0.4 mg / L and 6.03 mg / L during the manual control stage, the dissolved oxygen in the aerobic zone was generally maintained in the range of 1.56 mg / L to 2.87 mg / L during the period when the intelligent aeration system server implemented control, and more than 96.79% of the dissolved oxygen control values ​​had an error of less than 0.3 mg / L from the actual values, achieving the goal of following the changes in the influent water load.

[0089] Regarding water quality assurance and energy-saving effects, the parameters of various water quality indicators at the inlet and outlet of the wastewater treatment plant during the test period are shown in Table 1.

[0090] Table 1. Parameters of various water quality indicators at the influent and effluent of the wastewater treatment plant during the test period

[0091] Table 1 shows that the influent water quality fluctuated during the experiment, with the maximum fluctuation in chemical oxygen demand (COD) exceeding 160 mg / L and the maximum total nitrogen reaching 51.24 mg / L. Under the control of the intelligent aeration system server, the various indicators of the effluent remained at relatively stable levels. Specifically, the average COD of the effluent was 13.05 mg / L, the average ammonia nitrogen was 1.97 mg / L, and both total nitrogen and total phosphorus were better than the national Class A discharge standard (standard limits in Table 1). Meanwhile, the actual aeration energy consumption per ton of water under the control of the intelligent aeration system server was reduced by approximately 20% compared to the manual control stage, improving the operational economy of the wastewater treatment plant.

[0092] In summary, this embodiment verifies the practical effect of the coupled architecture of ASM2D mechanism model and LSTM algorithm.

[0093] First, the intelligent aeration system server calculates process water quality parameters through a mechanistic model, which reduces the system's reliance on some process water quality sensors to a certain extent. Second, the system can correct the influent ratio and dynamic parameters during continuous operation through the self-calibration of mechanism model parameters and the self-learning mechanism of LSTM algorithm, which is conducive to maintaining the accuracy of the algorithm in long-term operation. Third, the architecture utilizes a mechanistic model to perform pre-calculations for specific processes (such as oxidation ditches and AAO), while the LSTM algorithm focuses on dissolved oxygen trend prediction, indicating that the system is suitable for wastewater treatment plants using the activated sludge process. Experimental results show that, while ensuring that the effluent quality meets standards, the system can achieve stable control of dissolved oxygen and reduce aeration energy consumption.

[0094] During system operation, the actual operating conditions changed abruptly, with the sludge age increasing sharply from 12 days to 18 days. This caused the absolute error between the effluent ammonia nitrogen value simulated by the ASM2D mechanism model and the measured value to exceed 10%, thus formally triggering the system's parameter self-calibration procedure.

[0095] The system first identifies sludge age variation as the primary source of interference based on error attribution analysis, and then identifies high-sensitivity parameters according to a parameter sensitivity ranking table. The system employs an incremental step-size control strategy, prioritizing the adjustment of the maximum specific growth rate of nitrifying bacteria from the default value of 0.45 per day to 0.41 per day, followed by adjusting the denitrification rate from 0.15 to 0.14.

[0096] After the above calibration, the model's ammonia nitrogen prediction error decreased from 21.7% to 5.3%, and the aeration overshoot time was shortened from 8.5 hours to 2.1 hours, which helps to correct control deviations caused by operating condition drift.

Claims

1. An ASM model and AI based precise aeration system, characterized in that, include: The front-end water quality sensing equipment is configured to collect influent water volume data, influent water quality data, and environmental parameters during the wastewater treatment process; The PLC control system is configured to receive the influent water volume data, the influent water quality data, and the environmental parameters, and to execute control commands. The sewage treatment plant's supervisory control and data transmission system is configured to transmit data between the PLC control system and the intelligent aeration system server. The intelligent aeration system server runs an ASM2D mechanism model and an LSTM algorithm. It is configured to combine carbon source addition data and wastewater treatment operation status data, simulate online using the ASM2D mechanism model and output intermediate state parameters that characterize the micro-state of the biochemical system, and use the LSTM algorithm to receive the intermediate state parameters as input features to predict the corresponding aeration volume control command. The blower unit and related equipment are connected to the PLC control system and configured to respond to the control commands to adjust the aeration volume.

2. The accurate aeration system based on ASM model and AI according to claim 1, characterized in that, The intelligent aeration system server is configured to write back the predicted aeration volume data to the sewage treatment plant's upper computer monitoring system. The sewage treatment plant's upper computer monitoring system is configured to convert the aeration volume data into control commands and send them to the PLC control system. The PLC control system is configured to adjust the working status of the relevant equipment of the blower unit according to the control commands to execute intelligent aeration.

3. The accurate aeration system based on ASM model and AI according to claim 1, characterized in that, The front-end water quality sensing equipment includes water quality monitoring instruments, water quantity monitoring instruments, and environmental parameter sensors. The water quality monitoring instruments include an online COD analyzer and an online ammonia nitrogen analyzer. The online COD analyzer is used to detect chemical oxygen demand, and the online ammonia nitrogen analyzer is used to detect ammonia nitrogen concentration. The water volume monitoring instrument is installed at least once in the main inlet pipe and the open inlet channel. The water volume monitoring instrument is selected from at least one of electromagnetic flow meters and ultrasonic open channel flow meters. The water volume monitoring instrument is used to provide continuous instantaneous flow data. The environmental parameter sensors include a dissolved oxygen sensor, a temperature sensor, a pH meter, and a pressure sensor. The real-time dissolved oxygen concentration output by the dissolved oxygen sensor is used as the controlled variable for closed-loop feedback control of the system.

4. The accurate aeration system based on ASM model and AI according to claim 1, characterized in that, The intelligent aeration system server constructs a biochemical reaction simulation model based on the ASM2D mechanism model. The intelligent aeration system server divides the wastewater treatment biochemical unit into multiple completely mixed reaction tanks, numbers the tanks sequentially according to the influent sequence, and defines the influent of any reaction tank as consisting of three parts: front-end influent, sludge return, and mixed liquor return. The intelligent aeration system server establishes biochemical reaction kinetic equations based on the principle of mass balance. The principle of mass balance is as follows: The rate of change of component mass in any reaction tank over time is equal to the sum of the component mass rate flowing into any reaction tank from the previous reaction tank, the component mass rate brought into any reaction tank by sludge return and mixed liquor return, minus the component mass rate flowing out of any reaction tank, and plus the component mass rate generated by the biochemical reaction in any reaction tank.

5. The accurate aeration system based on ASM model and AI according to claim 4, characterized in that, The intelligent aeration system server calculates the theoretical total oxygen demand by establishing a dissolved oxygen mass balance equation based on the principle of dissolved oxygen balance. The dissolved oxygen balance principle is characterized as follows: The rate of change of dissolved oxygen concentration in the reaction tank is equal to the difference between the dissolved oxygen flux brought in by the upstream water and the dissolved oxygen flux flowing out of the reaction tank, minus the oxygen consumption rate, and plus the oxygen transfer rate. The intelligent aeration system server sets the rate of change of dissolved oxygen concentration to zero, calculates the theoretical total oxygen demand required to maintain the stability of dissolved oxygen in the aerobic tank by integral calculation, and calculates the theoretical air supply of the blower by combining parameters such as oxygen transfer efficiency in wastewater, water quality, temperature, pressure and immersion depth of aeration heads.

6. The accurate aeration system based on ASM model and AI according to claim 1, characterized in that, The intelligent aeration system server constructs an aeration volume prediction algorithm based on the LSTM algorithm. It takes the influent flow rate, aerobic tank water quality parameters including chemical oxygen demand and nitrogen and phosphorus indicators, sludge concentration, environmental parameters, dissolved oxygen at the previous moment and equipment status parameters as inputs, and the predicted dissolved oxygen value of the aerobic zone as output. The intelligent aeration system server executes the reverse optimization logic as follows: Using the theoretical air supply calculated by the ASM2D mechanism model as the initial value, the input aeration air volume is iteratively adjusted through the LSTM algorithm until the output predicted dissolved oxygen and the target value meet the preset control accuracy, thus determining the target aeration volume.

7. The accurate aeration system based on ASM model and AI according to claim 1, characterized in that, The coupled operating logic executed by the intelligent aeration system server includes a model training and construction phase and an online real-time control phase; In the model training and construction phase, a training database is generated, and the LSTM algorithm is trained under supervision to establish a mapping relationship between input features and aeration control quantities. During the online real-time control phase, the ASM2D mechanism model is run to generate intermediate parameters. The LSTM algorithm uses the real-time monitoring data and the intermediate parameters as joint inputs to directly generate aeration control commands that drive the blower unit.

8. The accurate aeration system based on ASM model and AI according to claim 4, characterized in that, The intelligent aeration system server is equipped with self-calibration logic for physical model parameters: When the error between the simulated effluent quality and the measured data from the ASM2D mechanism model continues to exceed the preset deviation threshold, the intelligent aeration system server increases all stoichiometric coefficients and kinetic parameters by a preset ratio and calculates the sensitivity of each parameter based on the simulation results. The intelligent aeration system server uses a sensitivity evaluation method based on the ASM model parameters to adjust each parameter sequentially according to its sensitivity, until the error between the simulated effluent water quality and the water quality monitored by the front-end water quality sensor is less than the preset deviation threshold.

9. The accurate aeration system based on ASM model and AI according to claim 1, characterized in that, The intelligent aeration system server is configured with AI algorithm self-learning logic, specifically as follows: When the intelligent aeration system server is running, it records the dissolved oxygen and corresponding aeration air volume in the aerobic tank, as well as the water quality simulated by the ASM2D mechanism model and the environmental parameters monitored by the front-end water quality sensing device. The above data are then unified into the same time series to form an array and stored in the database of the algorithm model. The intelligent aeration system server periodically trains and updates the model, and uses the new trained model to predict subsequent aeration volumes.

10. The accurate aeration system based on ASM model and AI according to claim 1, characterized in that, The PLC control system includes a central processing unit, an analog input module, an analog output module, and a communication module; the blower unit related equipment includes a blower unit, an electric regulating valve, and an air flow meter, and the blower unit is equipped with a variable frequency speed controller.

Images