Aeration control method, system and equipment for sewage treatment process and storage medium

Abstract

The invention provides an aeration control method for a sewage treatment process. The aeration control method comprises the following steps: acquiring real-time process parameters of a sewage treatment process; inputting the real-time process parameters into a pre-trained microbial oxygen consumption prediction model to obtain a future oxygen consumption prediction value; inputting the real-time process parameters and the future oxygen consumption predicted value into a pre-trained decision control model, wherein the decision control model generates an aeration control instruction based on a multi-objective optimization function; and sending an aeration control instruction to control the aeration equipment. According to the method, the obtained real-time process parameters are input into the microorganism oxygen consumption prediction model to obtain the future oxygen consumption prediction value, and prospective prediction is carried out on the future oxygen consumption; an aeration control instruction is generated based on the optimization target through the decision control model, and intelligent decision making is achieved; and the aeration equipment is controlled by sending an instruction to complete a control closed loop. The technical problems that in a traditional aeration control method, the control logic is disjointed with the real aerobic demand of microorganisms, and the control precision is insufficient are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, and in particular to an aeration control system, equipment, and storage medium for a wastewater treatment process. Background Technology

[0002] In wastewater treatment processes, the aeration system is the core component of aerobic biological treatment. Its main function is to provide sufficient dissolved oxygen for activated sludge microorganisms to meet the oxygen requirements of organic matter degradation and nitrification. However, the aeration system is also the most energy-consuming unit in a wastewater treatment plant, typically accounting for more than 50% of the plant's total electricity consumption.

[0003] Currently, the mainstream aeration control method mainly adopts PID control based on a fixed dissolved oxygen (DO) setpoint. However, this control method has the following technical drawbacks: First, the control logic is disconnected from the essence of the process. The PID controller aims to maintain the dissolved oxygen concentration at a manually set fixed value, but the actual oxygen demand of microorganisms fluctuates drastically with factors such as influent pollutant load, water temperature, and sludge activity. A fixed DO setpoint leads to over-aeration at low loads, resulting in wasted energy, and insufficient aeration at high loads, leading to incomplete nitrification and excessive ammonia nitrogen in the effluent. Second, the control suffers from a severe hysteresis effect. DO itself has control challenges such as hysteresis, complexity, and nonlinearity. PID control adjusts the aeration rate according to the current DO value, but the current DO value reflects the adjustment result from 15 minutes ago. This hysteresis causes the DO value to fluctuate continuously, resulting in generally low control accuracy. Third, it is difficult to achieve coordinated control of multiple devices. In traditional control systems, the frequency conversion regulation of the blower and the opening control of various valves on the air pipeline are often disconnected, frequently leading to the blower operating at inefficient operating points or unreasonable air volume distribution in different aeration zones, resulting in low overall aeration efficiency. Fourth, it relies on expensive hardware and lacks fault tolerance. Specialized instruments for online measurement of activated sludge oxygen uptake rate (OUR) are extremely expensive, making them unusable for most wastewater treatment plants. Furthermore, when sensors malfunction, traditional systems often directly trigger alarms and shut down, lacking intelligent fault tolerance mechanisms.

[0004] Therefore, there is an urgent need to provide an aeration control method that enables multi-device collaborative control based on accurate perception of the actual oxygen demand of microorganisms, while also possessing intelligent fault tolerance capabilities. Summary of the Invention

[0005] This invention achieves multi-dimensional state perception by acquiring real-time process parameters, laying a data foundation for precise control; it obtains predicted future oxygen consumption values ​​through a microbial oxygen consumption prediction model, enabling forward-looking prediction of future oxygen consumption; it generates aeration control commands based on optimization objectives through a decision control model, achieving intelligent decision-making; and it completes the control loop by sending commands to control the aeration equipment, ensuring precise execution of the strategy. This invention effectively solves the technical problems of the disconnect between control logic and the actual oxygen demand of microorganisms and the lack of foresight in traditional aeration control methods.

[0006] The purpose of this invention is to provide an aeration control method for a wastewater treatment process; The technical solution provided by this invention is as follows: An aeration control method for a wastewater treatment process includes: Obtain real-time process parameters for wastewater treatment; The real-time process parameters are input into a pre-trained microbial oxygen consumption prediction model to obtain the predicted future oxygen consumption value. Real-time process parameters and predicted future oxygen consumption are input into a pre-trained decision control model to obtain aeration control commands. The decision control model is trained using a multi-objective optimization function as the training objective. Send the aeration control command to control the aeration equipment.

[0007] Preferably, the real-time process parameters include influent water quality data, biological tank status data, aeration equipment operating status data, and time-of-use electricity price data.

[0008] Preferably, the microbial oxygen consumption prediction model is a deep learning model based on a causal discovery algorithm; The step of inputting the real-time process parameters into a pre-trained microbial oxygen consumption prediction model to obtain future oxygen consumption prediction values ​​includes: The causal discovery algorithm in the deep learning model is used to identify key process features that are causally related to the microbial oxygen consumption rate in the real-time process parameters from the pre-generated causal map. Based on the aforementioned key process characteristics, the real-time oxygen consumption rate value was estimated. Based on the real-time oxygen consumption rate value, the predicted future oxygen consumption value is obtained.

[0009] Preferably, the decision control model is a policy network based on a deep reinforcement learning algorithm; The multi-objective optimization function is a penalty function, which is composed of a weighted sum of an energy consumption penalty term, an oxygen demand deviation penalty term, and a stability penalty term. The energy consumption penalty term is calculated based on the time-of-use electricity price in the real-time process parameters. The step of inputting the real-time process parameters and the predicted future oxygen consumption into a pre-trained decision control model to obtain aeration control commands includes: The real-time process parameters and the predicted future oxygen consumption are input into the strategy network, and the strategy network outputs an aeration control command that minimizes the penalty function value.

[0010] Preferably, the aeration control commands include: blower group control commands, valve opening control commands, and main pipe pressure control commands.

[0011] Preferably, the step of inputting the real-time process parameters into a pre-trained microbial oxygen consumption prediction model to obtain future oxygen consumption prediction values ​​further includes: Real-time monitoring of the sensor status corresponding to the key process features; When a sensor malfunction is detected, alternative process features that are causally related to the key process features corresponding to the malfunctioning sensor are determined based on the causal graph. Based on the characteristics of the alternative process, the real-time oxygen consumption rate was estimated. Based on the real-time oxygen consumption rate value, the predicted future oxygen consumption value is obtained.

[0012] Preferably, after sending the aeration control command to control the aeration equipment, the method further includes: The real-time process parameters after the aeration control command is executed are obtained to form empirical data. The microbial oxygen consumption prediction model and / or the decision control model are iteratively trained based on the empirical data to achieve continuous optimization of the control strategy.

[0013] The second objective of this invention is to provide an aeration control system for a wastewater treatment process; The technical solution provided by this invention is as follows: An aeration control system for a wastewater treatment process includes: The acquisition module is used to acquire real-time process parameters of the wastewater treatment process; The prediction module is used to input the real-time process parameters into a pre-trained microbial oxygen consumption prediction model to obtain the predicted value of future oxygen consumption. The instruction generation module is used to input real-time process parameters and future oxygen consumption prediction values ​​into a pre-trained decision control model to obtain aeration control instructions. The decision control model is trained with a multi-objective optimization function as the training objective. The control module is used to send the aeration control commands to control the aeration equipment.

[0014] The third objective of this invention is to provide a computer device; The technical solution provided by this invention is as follows: A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of any one of the aeration control methods described above.

[0015] A fourth objective of this invention is to provide a computer-readable storage medium; The technical solution provided by this invention is as follows: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any one of the aeration control methods.

[0016] This invention provides an aeration control method for a wastewater treatment process, comprising: acquiring real-time process parameters of the wastewater treatment process; inputting the real-time process parameters into a pre-trained microbial oxygen consumption prediction model to obtain predicted future oxygen consumption values; inputting the real-time process parameters and the predicted future oxygen consumption values ​​into a pre-trained decision control model, which generates aeration control commands based on a multi-objective optimization function; and sending the aeration control commands to control the aeration equipment. This invention achieves intelligent decision-making by acquiring real-time process parameters and inputting them into a microbial oxygen consumption prediction model to obtain predicted future oxygen consumption values; generating aeration control commands based on optimization objectives through the decision control model; and completing the control loop by sending commands to control the aeration equipment. This invention effectively solves the technical problem of insufficient control accuracy and disconnect between control logic and the actual oxygen demand of microorganisms in traditional aeration control methods.

[0017] The present invention also provides an aeration control system for a wastewater treatment process. Since this system and the aeration control method solve the same technical problem and belong to the same technical concept, they should have the same beneficial effects, and will not be described in detail here. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of an aeration control method for a wastewater treatment process according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the aeration control system of a wastewater treatment process according to an embodiment of the present invention; Figure 3This is an internal structural diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

[0020] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] like Figure 1 As shown, this embodiment of the invention provides an aeration control method for a wastewater treatment process, comprising: S1. Obtain real-time process parameters for the wastewater treatment process; In this embodiment, the wastewater treatment plant deploys various online monitoring instruments, including: an online chemical oxygen demand (COD) analyzer, an online ammonia nitrogen analyzer, and an electromagnetic flow meter at the inlet; a dissolved oxygen (DO) sensor, a mixed liquor suspended solids (MLSS) sensor, a temperature sensor, and a pH meter in the biological treatment tank; and a pressure sensor, a flow meter, and opening feedback devices for each regulating valve at the blower outlet. Step S1 involves acquiring real-time process parameters of the wastewater treatment process collected by the aforementioned monitoring instruments. This step achieves comprehensive real-time perception of the wastewater treatment process, providing a high-quality data foundation for precise control and avoiding control deviations caused by incomplete information or data distortion.

[0022] Preferably, the real-time process parameters include influent water quality data, biological tank status data, aeration equipment operating status data, and time-of-use electricity price data.

[0023] In practical applications, influent water quality data includes, but is not limited to: influent flow rate (Q_in), influent COD concentration (COD_in), influent ammonia nitrogen concentration (NH3-N_in), and influent total nitrogen concentration (TN_in). Biological tank status data includes, but is not limited to: dissolved oxygen concentration (DO_1, DO_2, …), oxidation-reduction potential (ORP), pH value, temperature, mixed liquor suspended solids concentration (MLSS), and ammonia nitrogen (NH3-N_aerobic_end) and nitrate concentration (NO3-N_aerobic_end) at the end of the aerobic tank. Aeration equipment operating status data includes, but is not limited to: the number of blowers in operation, frequency / current of each blower, outlet pressure and flow rate, main pipe pressure, and opening feedback of regulating valves in each aeration branch. Time-of-use electricity price data includes electricity price information for different time periods, such as peak hours, normal hours, and valley hours.

[0024] In practical application, step S1 includes: S11. Perform data cleaning on the real-time process parameters. Filter outliers and intelligently imput missing values ​​in the real-time process parameter data to ensure data reliability.

[0025] S12. Perform vector processing on the cleaned real-time process parameters to obtain a real-time process parameter feature vector. Integrate the above data for each control cycle (e.g., 1 minute) into a real-time process parameter vector S(t) describing the overall state of the system at time t. For example: S(t) = [Q_in, real-time nitrification load, DO_1..n, ORP, MLSS, NH3-N_aerobic_end, T, fan frequency, main pipe pressure, valve opening, electricity price period, ...].

[0026] S2. Input the real-time process parameters into the pre-trained microbial oxygen consumption prediction model to obtain the predicted value of future oxygen consumption. In this embodiment, a pre-trained microbial oxygen consumption prediction model is deployed. This model, built on a deep neural network, takes real-time process parameter time-series data from the past period (e.g., 1 hour) as input and outputs predicted microbial oxygen consumption values ​​for the next period (e.g., the next 2 hours). During the model training phase, historical operating data from the wastewater treatment plant is used to learn the nonlinear mapping relationship between parameters such as influent water quality, sludge concentration, and temperature and oxygen consumption offline. After training, the model is solidified and deployed to an edge server. In actual operation, at regular intervals (e.g., 1 minute), the real-time parameter vector obtained in step S1 is input into the model, and the model calculates and outputs predicted oxygen consumption values ​​for the next time period (e.g., the next two hours). It should be noted that microbial oxygen consumption (usually expressed as oxygen consumption rate OUR) is a direct reflection of the metabolic activity of activated sludge and a fundamental basis for determining aeration oxygen demand. Predictive models trained on historical data can uncover the complex relationships between multiple parameters and oxygen consumption, thereby enabling forward-looking predictions of future oxygen demand trends and transforming control from "post-event adjustment" to "pre-event planning." This step replaces expensive hardware equipment for measuring microbial oxygen consumption with software algorithms, obtaining key information on microbial oxygen demand at low cost, while providing predictions of future trends. This provides forward-looking input for subsequent optimization decisions, effectively overcoming the oscillations and overshoots caused by the lag in traditional control methods.

[0027] Preferably, the microbial oxygen consumption prediction model is a deep learning model based on a causal discovery algorithm; Real-time process parameters are input into a pre-trained microbial oxygen consumption prediction model to obtain future oxygen consumption predictions, including: S21. Using the causal discovery algorithm in the deep learning model, identify key process features in the real-time process parameters that are causally related to the microbial oxygen consumption rate from the pre-generated causal map. In this embodiment, a causal graph describing the causal relationships between various process parameters is constructed in advance based on the historical operating data of the wastewater treatment plant using causal discovery algorithms (such as PC algorithm, GES algorithm, or score-based search method). This graph is represented in the form of a directed graph, for example: influent ammonia nitrogen concentration → nitrification oxygen demand, influent COD concentration → carbon oxidation oxygen demand, MLSS → endogenous respiration oxygen demand, water temperature → regulates the rates of all biological reactions, etc.

[0028] As a specific implementation method, real-time process parameters also include: action parameters, exogenous variables, and constants. Action parameters record the control commands for the current cycle, including: internal reflux ratio setting, external reflux ratio setting, aeration fan frequency setting, carbon source dosage setting, etc. Exogenous variables and constants are parameters in the wastewater treatment system that are unaffected by the actions; that is, the values ​​of exogenous variables and constants do not change with changes in action data. For example, exogenous variables can be external parameters of the wastewater treatment system, such as meteorological parameters like ambient temperature and ambient light. Constants can be fixed parameters within the wastewater treatment system, such as peak influent flow rate and peak effluent flow rate. A causal graph is used to formally describe the complex causal relationships between variables in the wastewater treatment process. Therefore, this embodiment constructs state nodes based on influent water quality data and biological tank status data, action nodes based on action parameters, and intermediate nodes based on exogenous variables and constants. A causal graph, i.e., a directed graph, is constructed based on state nodes, action nodes, and intermediate nodes to represent the entire wastewater treatment system. Nodes with mutual influence relationships are connected by graph edges, representing the dependencies and causal relationships between variables. For example, the current "aeration rate" action data and the current "dissolved oxygen concentration in the biochemical reactor" state data jointly determine the "dissolved oxygen concentration in the biochemical reactor" state data at the next moment. Decision nodes and inference nodes are determined based on the node relationships in the causal graph. Specifically, when the parent nodes of a node determined based on the node relationships in the causal graph are all state nodes and / or intermediate nodes, the node is determined as a decision node; when the parent nodes of a node determined based on the node relationships in the causal graph include action nodes, the node is determined as an inference node. This embodiment can assign an independent neural network to each decision node and inference node.

[0029] In actual operation, after obtaining the real-time process parameter vector at the current moment in step S1, the model first loads the pre-generated causal map. Then, through the built-in causal feature discovery mechanism (such as backdoor adjustment, frontdoor adjustment, or instrumental variable method based on causal effect estimation), the causal effect strength between the current process parameters and the microbial oxygen consumption rate is calculated. Only those parameters that are determined to have a direct or indirect causal effect on the oxygen consumption rate are identified as "key process features".

[0030] It's important to note that traditional deep learning models (such as standard LSTM) primarily rely on statistical correlation for prediction. When extreme conditions not covered by the training data occur, the model is prone to making incorrect predictions due to "spurious correlations." The introduction of causal discovery algorithms, however, enables the model to distinguish between "cause" and "effect," truly understanding which factors are the root causes driving changes in oxygen consumption. The pre-generated causal map focuses on truly important features in each prediction, eliminating irrelevant noise.

[0031] S22. Based on key process characteristics, the real-time oxygen consumption rate is estimated. In this embodiment, after identifying key process features, the model enters the oxygen consumption rate estimation stage. The model employs a structured deep learning architecture whose network structure matches the causal relationships revealed by the causal graph in S21. For example, the model contains multiple parallel sub-network modules: one sub-network specifically handles key features related to nitrification (such as influent ammonia nitrogen load) and outputs the contribution value of nitrification oxygen demand; another sub-network handles key features related to carbon oxidation (such as influent COD load) and outputs the contribution value of carbon oxidation oxygen demand; a third sub-network handles key features related to endogenous respiration (such as MLSS) and outputs the contribution value of endogenous respiration oxygen demand. Finally, a fusion layer weights and sums these contributions and introduces a regulating factor (such as water temperature) to obtain the final real-time oxygen consumption rate estimate.

[0032] Designing a network structure based on causal graphs essentially involves injecting domain knowledge into the model in the form of network topology, guiding the model to learn feature mapping relationships that conform to physical laws. This approach retains the powerful nonlinear fitting capabilities of deep learning while also endowing the model with interpretability and logical consistency. This step enables real-time and accurate estimation of microbial oxygen consumption rates.

[0033] S23. Based on the real-time oxygen consumption rate, the predicted value of future oxygen consumption is obtained.

[0034] In this embodiment, after obtaining the real-time oxygen consumption rate value at the current moment, the model further combines it with the historical oxygen consumption rate sequence over a past period (e.g., the past hour) and inputs it into a time-series prediction module. As one implementation, this module employs an Encoder-Decoder architecture. The Encoder part processes the historical sequence, extracting time-series features and trends; the Decoder part, based on the Encoder's output and the current value, generates a predicted oxygen consumption sequence for a future period (e.g., the next two hours). The prediction results are output at equal time intervals (e.g., every 15 minutes), forming a complete characterization of future oxygen demand trends.

[0035] During the prediction process, the model also outputs a confidence interval for each predicted value to characterize the uncertainty of the prediction. When the confidence interval is too wide (i.e., the prediction uncertainty is high), the system can adopt a more conservative control strategy in subsequent decisions.

[0036] In practical applications, the deep learning model based on the causal discovery algorithm can be implemented using various neural network architectures. For example, in the causal feature discovery stage (S21), a graph attention network (GAT) can be used. A pre-generated causal graph is used as input to the graph structure, and the causal attention weights of each process parameter node to the oxygen consumption rate target node are calculated through the graph attention mechanism, thereby identifying key features. In the real-time oxygen consumption rate estimation stage (S22), a multi-layer stacked long short-term memory network (LSTM) can be used. Historical time-series data of the selected key features are input into the network, and its gating mechanism is used to capture the dynamic changes in microbial activity, outputting the current oxygen consumption rate estimate. In the future oxygen consumption prediction stage (S23), an encoder-decoder architecture based on gated recurrent units (GRUs) can be used. The encoder reads the historical oxygen consumption rate sequence of the past hour, and the decoder gradually generates the oxygen consumption prediction sequence for the next two hours. Alternatively, a temporal convolutional network (TCN) can be used to process time-series data in parallel to improve the response speed to sudden shock loads. Those skilled in the art can choose one or more combinations of the above models for training and deployment, depending on the actual data scale and computing power limitations.

[0037] Preferably, inputting real-time process parameters into a pre-trained microbial oxygen consumption prediction model to obtain future oxygen consumption prediction values ​​further includes: S24. Real-time monitoring of sensor status corresponding to key process characteristics; S25. When a sensor malfunction is detected, based on the causal graph, identify alternative process features that are causally related to the key process features corresponding to the malfunctioning sensor. S26. Based on the characteristics of the alternative process, the real-time oxygen consumption rate is estimated. S27. Based on the real-time oxygen consumption rate, the predicted value of future oxygen consumption is obtained.

[0038] Step S24 establishes a real-time sensor health status monitoring thread, which polls all sensors involved in key process features in real time. The monitored content includes: whether the sensor reading is within a reasonable range (e.g., 0-10 mg / L for DO sensor), whether the rate of change of the reading exceeds the physical limit (e.g., a sudden change of more than 1 mg / L in DO within 1 second is usually abnormal), the sensor self-diagnostic status flag (e.g., most smart sensors have self-testing functions), and the magnitude of the residual between the reading and the model's expected value.

[0039] Once S24 determines that a sensor has malfunctioned, it immediately queries a pre-generated causal graph. The causal graph records not only the type of causal relationship between various process parameters (direct causality, indirect causality, reverse causality), but also the strength of the causal relationship (such as the causal effect coefficient) and time delay information (such as the effect of changes in influent ammonia nitrogen on oxygen consumption rate with a lag of approximately 15 minutes). Based on the causal graph, the system executes the following logic to determine alternative process features: First, it locates the process feature node corresponding to the faulty sensor in the causal graph (such as the "influent ammonia nitrogen" node). Second, it searches for other nodes causally related to this node, especially those predecessor nodes that act as "causes" pointing to this node (such as "industrial wastewater discharge ratio" and "influent flow rate"), as well as successor nodes that act as "causes" pointing to this node (such as "nitrification oxygen demand" and "ammonia nitrogen at the end of the aerobic tank"). Then, based on the current sensor status of each candidate node, it selects nodes with normal status as candidate alternative features. Finally, based on the causal effect strength and time delay information, it sorts the candidate alternative features and selects the optimal alternative process feature. For example, if the influent ammonia nitrogen sensor malfunctions, but the cause-effect diagram shows that "influent COD" and "influent ammonia nitrogen" have a common pollution source (causal relationship), and the COD sensor is functioning normally, then the system selects "influent COD" as the alternative process feature.

[0040] After determining the characteristics of the alternative process, the model enters a fault-tolerant mode for estimating the oxygen demand rate. Step S26, the estimation process, is still based on a deep learning model, but the input data is adjusted: readings from faulty sensors are marked as unavailable, and their corresponding input channels are filled with real-time data from the alternative process characteristics. Internally, due to the use of a structured design based on causal graphs (e.g., parallel nitrification and carbonization sub-networks), when influent ammonia nitrogen data is unavailable, the input to the nitrification sub-network can be flexibly switched to other variables causally related to the nitrification reaction. For example: Option A: Using influent COD and influent flow rate, a lightweight auxiliary estimation model is used to first estimate the current ammonia nitrogen load, and then input into the nitrification sub-network. Option B: Directly bypassing the nitrification sub-network, the ammonia nitrogen concentration at the end of the aerobic tank (as the "effect" of nitrification) and DO change rate are used to infer the nitrification oxygen demand. Option C: Reducing the weight of the nitrification sub-network, relying more on the contributions of the carbonization and endogenous respiration sub-networks, while performing rough compensation based on the historical nitrification proportion for the same period. After obtaining the real-time oxygen consumption rate estimate in fault-tolerant mode, the future oxygen consumption prediction process continues.

[0041] This application introduces an intelligent fault-tolerant mechanism based on causal graphs. When a sensor fails, instead of simply shutting down the machine or using interpolation from adjacent sensors, it finds the most reasonable alternative variable based on causal relationships to obtain the real-time oxygen consumption rate value and the predicted value of future oxygen consumption. This preserves the forward-looking control capability of aeration control and avoids degradation to manual control or fixed value control due to sensor failure.

[0042] S3. Input the real-time process parameters and the predicted future oxygen consumption into the pre-trained decision control model to obtain the aeration control command, wherein the decision control model is trained with a multi-objective optimization function as the training objective. In this embodiment, a pre-trained decision control model is also deployed. This model is also based on a deep neural network, and its input consists of two parts: the real-time process parameter vector at the current moment and the predicted future oxygen consumption value. During the training phase, the decision control model employs reinforcement learning or supervised learning methods to minimize aeration energy consumption while ensuring the effluent quality meets standards. This means minimizing the long-term comprehensive operating cost while satisfying all preset constraints. In practical application, the preset constraints include: Process constraints: DO_min ≤ DO(i) ≤ DO_max (safe range), where DO_min and DO_max are the minimum dissolved oxygen concentrations, and their values ​​are determined based on historical experience; NH3-N_pred(i) ≤ national emission standards, where NH3-N_pred(i) is the predicted ammonia nitrogen value for the i-th cycle. Equipment constraints: fan frequency, pressure, upper and lower limits of valve opening; fan surge protection curve. Coupling constraints: the main pipe pressure must meet the flow requirements of each branch valve opening.

[0043] The decision control model comprehensively considers current operating conditions and future oxygen demand changes, balancing multiple objectives such as energy consumption, oxygen demand deviation, and equipment stability to generate a set of control actions that optimize overall system performance. In practical applications, the output of the decision control model is a multi-dimensional control command vector, including the frequency setpoints of multiple blowers, the main pipe pressure setpoint, and the opening setpoints of the regulating valves in each aeration branch. This model can handle complex control challenges such as multivariate coupling, nonlinearity, and time-varying conditions. This step realizes the transformation from single-point setpoint control to multi-objective intelligent collaborative control, enabling the aeration system to dynamically adapt to load changes, automatically balance energy consumption and treatment effect, avoid the lag and inaccuracy of manual adjustment, and significantly improve control quality.

[0044] Preferably, the aeration control commands include: blower group control commands, valve opening control commands, and main pipe pressure control commands.

[0045] Preferably, the decision control model is a policy network based on a deep reinforcement learning algorithm; The multi-objective optimization function is a penalty function, which is composed of a weighted sum of an energy consumption penalty term, an oxygen demand deviation penalty term, and a stability penalty term. The energy consumption penalty term is calculated based on the time-of-use electricity price in the real-time process parameters. The step of inputting the real-time process parameters and the predicted future oxygen consumption into a pre-trained decision control model to obtain aeration control commands includes: The real-time process parameters and the predicted future oxygen consumption are input into the strategy network, and the strategy network outputs an aeration control command that minimizes the penalty function value.

[0046] In this embodiment, the decision control model adopts a policy network architecture based on deep reinforcement learning. The policy network is a deep neural network whose input layer receives a state vector composed of real-time process parameters and predicted future oxygen consumption values, while the output layer generates multi-dimensional aeration control commands, including frequency setpoints for multiple blowers, main pipe pressure setpoints, and opening setpoints for regulating valves in each aeration branch. The training objective of the policy network is to learn an optimal mapping relationship from state to action, minimizing the accumulated penalty function value during interaction with the environment.

[0047] As a specific implementation method, the multi-objective optimization function expression is as follows: Min J = Σ[ω1* P_blower(U(i)) * f_electricity price(i) +ω2* (OUR_target(i) - OUR_pred(i))² + ω3* ΔU(i)² ], i=t to t+m-1 In the formula, T represents the current moment, which is the starting point for the optimization decision; m is the preset number of optimization steps; i is the index of each control cycle from the current time t to the future time t+m-1; P_blower(U(i)) represents the total power consumption of the blower group determined by the aeration control command U(i). U(i) is the aeration control command for the next m control cycles, such as the total air supply setting command for the blowers, the start / stop and frequency combination command for each blower, the main pipe pressure setting command, and the valve opening command for each aeration branch. In practical applications, when the model outputs a specific set of aeration control commands U(i) (e.g., start blower 1 at 40Hz; start blower 2 at 35Hz), the system can calculate the total power consumption P_blower under this combination based on the blower performance curves.

[0048] f_electricity price(i) is the time-of-use electricity price coefficient for the i-th period, and P_wind turbine(U(i)) * f_electricity price(i) constitutes the direct electricity cost for that period.

[0049] OUR_pred(i) is the predicted oxygen consumption rate for the i-th future period, output by the microbial oxygen consumption prediction model.

[0050] OUR_target(i): This is the theoretical oxygen demand target that is dynamically calculated based on the influent load and process requirements. This value can be pre-calculated based on historical experience data, or it can be calculated based on a corresponding neural network model trained on historical data.

[0051] (OUR_target(i) - OUR_pred(i))² represents the oxygen demand deviation, which is the difference between the "predicted oxygen demand" and the "theoretical oxygen demand". It drives the system to supply gas on demand and avoid supply and demand mismatch.

[0052] ΔU(i)² represents the sum of squares of the changes in the aeration control command relative to the control command of the previous cycle in the i-th control cycle. It is used to characterize the penalty for the change in the equipment action and to ensure smooth operation of the equipment.

[0053] ω1, ω2, and ω3 are weights determined through offline training to balance energy consumption, oxygen demand deviation, and equipment stability.

[0054] Meanwhile, the decision control model also needs to meet the following constraints during the training process: Process constraints: DO_min ≤ DO(i) ≤ DO_max (safe range); NH3-N_pred(i) ≤ national standard.

[0055] Equipment constraints: fan frequency, pressure, upper and lower limits of valve opening; fan surge protection curve.

[0056] Coupling constraint: The main pipe pressure must meet the flow requirements of each branch valve opening.

[0057] The training process employs policy gradient algorithms (such as the PPO or SAC algorithms) from reinforcement learning. In the initial training phase, the policy network is randomly initialized, and the output control commands may not be entirely reasonable. Through interaction with the environment, which can be offline historical data playback from an actual wastewater treatment plant or a high-fidelity digital twin simulation environment, the environment calculates an immediate penalty value based on the actual effect of each action after the policy network executes it, and feeds the new state back to the network. Based on this feedback, the policy network continuously adjusts its parameters, gradually learning which actions to take in which states to minimize the long-term accumulated penalty value.

[0058] The goal of the multi-objective optimization function is to minimize energy consumption, minimize oxygen demand deviation, and minimize equipment fluctuations. Therefore, the optimization objectives are defined as penalty terms, with smaller values ​​indicating better control performance. The penalty function consists of a weighted sum of three parts: the energy consumption penalty term characterizes the real-time operating cost of the aeration system, calculated based on the current total power consumption of the aeration system and weighted by the time-of-use electricity price for the current period. Time-of-use electricity price information is input into the model as part of the real-time process parameters, enabling the strategy network to learn an economical operating strategy of "saving electricity during peak hours and using energy rationally during off-peak hours." For example, during peak electricity price periods, the model tends to reduce aeration as much as possible while ensuring water quality; while during off-peak electricity price periods, aeration can be appropriately increased for cleaning membrane modules or maintaining sludge activity. The oxygen demand deviation penalty term characterizes the oxygen demand deviation, i.e., the difference between the "predicted oxygen demand" and the "theoretical oxygen demand," driving the system to supply air on demand and avoid supply-demand mismatch. The smaller the oxygen demand deviation, the smaller the penalty value, and vice versa. The stability penalty term characterizes the smoothness of equipment operation. Its calculation is based on the variation range of fan frequency, valve opening, and main pipe pressure. Frequent large-amplitude movements not only increase equipment wear and shorten service life, but also lead to system oscillation and decreased control quality. By introducing the stability penalty term, the policy network is guided to learn a smooth and gradual control strategy, avoiding unnecessary drastic adjustments.

[0059] S4. Send aeration control commands to control the aeration equipment.

[0060] In this embodiment, aeration control commands are sent to the blower control cabinet, the main pipe pressure regulating valve, and the actuators of each branch electric valve. Before being sent, the commands undergo safety checks (such as amplitude limit checks and rate of change limits) to ensure that equipment protection is not triggered or that dangerous operating conditions are not caused. Upon receiving the command, the actuator completes the action in the next control cycle, ensuring that the actual oxygen supply follows the predicted oxygen demand. This step completes the final step from decision-making to execution, ensuring that the optimization strategy is implemented effectively, achieving precise and coordinated control of the aeration equipment, and ultimately achieving the goal of on-demand oxygen supply.

[0061] Preferably, after sending the aeration control command to control the aeration equipment, the method further includes: S5. Obtain real-time process parameters after executing aeration control commands to form empirical data; In this embodiment, after each control cycle, a complete set of empirical data is automatically acquired and recorded. This set of data includes: the initial state before executing the control command (i.e., the real-time process parameters acquired in step S1 and the predicted future oxygen consumption value generated in step S2), the aeration control command output by the decision control model (i.e., the aeration control command generated in step S3), the new state reached by the system after executing the command (i.e., the real-time process parameters at the next moment), and the actual operating performance indicators generated during the control cycle (such as actual energy consumption, oxygen demand deviation, equipment movement amplitude, etc.).

[0062] S6. Iteratively train the microbial oxygen consumption prediction model and / or decision control model based on empirical data to achieve continuous optimization of the control strategy.

[0063] In practical applications, a cloud-edge collaborative architecture is adopted to achieve continuous model optimization. The cloud training platform regularly collects accumulated experience data from all edge servers, forming a continuously growing data pool. Based on this new data, the cloud platform initiates retraining or incremental training of the microbial oxygen consumption prediction model and the decision control model. After training is completed, the new model parameters undergo rigorous offline verification (such as verifying performance on the test set and simulating operation in a digital twin environment) before being distributed to the edge servers for hot updates. Upon receiving the new model, the edge servers can choose to seamlessly switch in the next control cycle, with no interruption to on-site operations throughout the entire process.

[0064] This invention achieves the self-evolution capability of the control strategy, specifically manifested in the following three aspects: First, continuously improving accuracy. As the operating time increases, the available training data for the predictive model becomes increasingly abundant, resulting in more comprehensive coverage of various operating conditions. Second, adaptive adaptation to changing operating conditions. When the influent water quality undergoes seasonal changes (such as high temperature and high load in summer, and low temperature and low load in winter), the system can automatically adapt. Third, discovery of better strategies. With the accumulation of empirical data, the decision control model has the opportunity to discover better control modes from a global perspective.

[0065] Compared to existing technologies, this application uses a microbial oxygen consumption prediction model to replace expensive online respirators, obtaining real oxygen demand information of microorganisms at low cost through software algorithms. Simultaneously, by predicting future oxygen consumption, aeration control is transformed from post-event adjustment to pre-event prediction, effectively overcoming the DO oscillation and overshoot problems caused by the lag in traditional control, and improving the system's ability to cope with shock loads. Furthermore, by employing a decision control model trained with a multi-objective optimization function as the training objective, it achieves coordinated optimization of multiple control objectives such as energy consumption, oxygen demand deviation, and stability. This solves the technical deficiency of traditional PID control, which can only maintain a fixed DO and cannot simultaneously consider multi-objective optimization, enabling the aeration system to automatically seek the optimal energy consumption operating point while ensuring effluent meets standards. This solution explicitly integrates time-of-use pricing into the optimization objective, enabling the decision control model to operate based on economical strategies of peak-hour energy conservation and off-peak energy utilization. Building upon traditional energy consumption optimization, it further refines energy cost control, significantly improving operational economics. Employing a causal discovery algorithm to identify the true drivers of oxygen consumption rate changes, rather than relying solely on statistical correlation, the model maintains predictive accuracy and robustness even in extreme conditions not covered by training data, addressing the insufficient generalization ability of purely data-driven models. By outputting multi-dimensional collaborative control commands covering blower groups, valve openings, and mains pressure through a policy network, it achieves end-to-end optimization from gas generation and distribution to pool-end gas supply, overcoming the limitations of isolated equipment operation and localized optimization in traditional control, resulting in a significant improvement in overall energy efficiency. Intelligent fault tolerance is achieved based on causal graphs, automatically and seamlessly switching to causally related alternative variables when sensors fail, maintaining continuous and stable system operation. This solves the problem of control failure due to single-point failures in traditional control systems, greatly improving the system's industrial reliability and availability. This solution constructs a closed-loop evolutionary mechanism of "perception-decision-execution-learning", which enables the model to continuously optimize itself as operational data accumulates, automatically adapt to long-term drift such as changes in influent water quality and equipment performance aging, and realizes the continuous evolution of control strategies and in-depth exploration of energy-saving potential.

[0066] like Figure 2 As shown, an embodiment of the present invention provides an aeration control system for a wastewater treatment process, comprising: The acquisition module is used to acquire real-time process parameters of the wastewater treatment process; The prediction module is used to input real-time process parameters into a pre-trained microbial oxygen consumption prediction model to obtain future oxygen consumption prediction values. The instruction generation module is used to input real-time process parameters and future oxygen consumption prediction values ​​into a pre-trained decision control model to obtain aeration control instructions. The decision control model is trained with a multi-objective optimization function as the training objective. The control module is used to send aeration control commands to control the aeration equipment.

[0067] Specific limitations regarding the aeration control system for a wastewater treatment process can be found in the above description of the aeration control method for a wastewater treatment process, and will not be repeated here. Each module in the aforementioned aeration control system for a wastewater treatment process can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0068] like Figure 3 As shown, in one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the following steps: Obtain real-time process parameters for wastewater treatment; Real-time process parameters are input into a pre-trained microbial oxygen consumption prediction model to obtain future oxygen consumption prediction values. Real-time process parameters and predicted future oxygen consumption are input into a pre-trained decision control model to obtain aeration control commands. The decision control model is trained using a multi-objective optimization function as the training objective. Send aeration control commands to control the aeration equipment.

[0069] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0070] In one embodiment, a computer-readable storage medium is provided, storing a computer program that, when executed by a processor, performs the following steps: Obtain real-time process parameters for wastewater treatment; Real-time process parameters are input into a pre-trained microbial oxygen consumption prediction model to obtain future oxygen consumption prediction values. Real-time process parameters and predicted future oxygen consumption are input into a pre-trained decision control model to obtain aeration control commands. The decision control model is trained using a multi-objective optimization function as the training objective. Send aeration control commands to control the aeration equipment.

[0071] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple modules or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, and can be electrical, mechanical, or other forms.

[0072] Furthermore, in the various embodiments of the present invention, each functional module can be fully integrated into a processor, or each module can be a separate device, or two or more modules can be integrated into a device; each functional module in the various embodiments of the present invention can be implemented in hardware or in the form of hardware plus software functional units.

[0073] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by program instructions and related hardware. The aforementioned program instructions can be stored in a computer-readable storage medium. When the program instructions are executed, they perform the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0074] It should be understood that the use of terms such as "system," "apparatus," "unit," and / or "module" in this application is only applicable to distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0075] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "a," and / or "the" are not specifically singular and may include the plural. Generally, the terms "comprising" and "including" only indicate the inclusion of expressly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements. An element defined by the phrase "comprising an..." does not exclude the presence of other identical elements in the process, method, product, or apparatus that includes the element.

[0076] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "a plurality of" or "several" means two or more, unless otherwise explicitly specified.

[0077] If a flowchart is used in this application, it is used to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0078] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An aeration control method for a wastewater treatment process, characterized in that, include: Obtain real-time process parameters for wastewater treatment; The real-time process parameters are input into a pre-trained microbial oxygen consumption prediction model to obtain the predicted future oxygen consumption value. The real-time process parameters and the predicted future oxygen consumption are input into a pre-trained decision control model to obtain aeration control commands. The decision control model is trained using a multi-objective optimization function as the training objective. Send the aeration control command to control the aeration equipment.

2. The aeration control method according to claim 1, characterized in that, The real-time process parameters include influent water quality data, biological tank status data, aeration equipment operating status data, and time-of-use electricity price data.

3. The aeration control method according to claim 2, characterized in that, The microbial oxygen consumption prediction model is a deep learning model based on a causal discovery algorithm. The step of inputting the real-time process parameters into a pre-trained microbial oxygen consumption prediction model to obtain future oxygen consumption prediction values ​​includes: The causal discovery algorithm in the deep learning model is used to identify key process features that are causally related to the microbial oxygen consumption rate in the real-time process parameters from the pre-generated causal map. Based on the aforementioned key process characteristics, the real-time oxygen consumption rate value was estimated. Based on the real-time oxygen consumption rate value, the predicted future oxygen consumption value is obtained.

4. The aeration control method according to claim 3, characterized in that, The decision control model is a policy network based on a deep reinforcement learning algorithm; The multi-objective optimization function is a penalty function, which is composed of a weighted sum of an energy consumption penalty term, an oxygen demand deviation penalty term, and a stability penalty term. The energy consumption penalty term is calculated based on the time-of-use electricity price in the real-time process parameters. The step of inputting the real-time process parameters and the predicted future oxygen consumption into a pre-trained decision control model to obtain aeration control commands includes: The real-time process parameters and the predicted future oxygen consumption are input into the strategy network, and the strategy network outputs an aeration control command that minimizes the penalty function value.

5. The aeration control method according to claim 4, characterized in that, The aeration control commands include: blower group control commands, valve opening control commands, and main pipe pressure control commands.

6. The aeration control method according to claim 3, characterized in that, The step of inputting the real-time process parameters into a pre-trained microbial oxygen consumption prediction model to obtain future oxygen consumption prediction values ​​also includes: Real-time monitoring of the sensor status corresponding to the key process features; When a sensor malfunction is detected, alternative process features that are causally related to the key process features corresponding to the malfunctioning sensor are determined based on the causal graph. Based on the characteristics of the alternative process, the real-time oxygen consumption rate was estimated. Based on the real-time oxygen consumption rate value, the predicted future oxygen consumption value is obtained.

7. The aeration control method according to claim 1, characterized in that, After sending the aeration control command to control the aeration equipment, the process further includes: The real-time process parameters after the aeration control command is executed are obtained to form empirical data. The microbial oxygen consumption prediction model and / or the decision control model are iteratively trained based on the empirical data to achieve continuous optimization of the control strategy.

8. An aeration control system for a wastewater treatment process, characterized in that, include: The acquisition module is used to acquire real-time process parameters of the wastewater treatment process; The prediction module is used to input the real-time process parameters into a pre-trained microbial oxygen consumption prediction model to obtain the predicted value of future oxygen consumption. The instruction generation module is used to input the real-time process parameters and the predicted future oxygen consumption into a pre-trained decision control model to obtain aeration control instructions, wherein the decision control model is trained with a multi-objective optimization function as the training objective; The control module is used to send the aeration control commands to control the aeration equipment.

9. A computer device, characterized in that, The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the aeration control method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the aeration control method as described in any one of claims 1-7.

Images