Knowledge-aided accurate aeration modeling method for urban sewage treatment

Abstract

The invention relates to the field of sewage treatment process control methods, in particular to an urban sewage treatment accurate aeration modeling method based on knowledge assistance, which comprises the following steps: establishing a data driving model according to acquired aeration process key data; based on an oxygen mass transfer-reaction coupling mechanism model, directional mining of key data in the aeration process is guided, knowledge extraction is carried out, new characteristic parameters associated with the aeration process are calculated, and a multi-dimensional characteristic matrix is formed; and jointly using the newly added multi-dimensional feature value and the original aeration process key data for training and optimization of a data-driven model to obtain an aeration model based on knowledge assistance. According to the method, knowledge and data are deeply fused, directional mining of data depth features is guided by the knowledge, and a good aeration simulation control effect is achieved based on defect data conditions.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment process control methods, specifically to a knowledge-assisted precise aeration modeling method for urban wastewater treatment. Background Technology

[0002] Currently, the most commonly used core process in wastewater treatment plants is the activated sludge process. Its principle is to utilize microorganisms in activated sludge to decompose organic matter in wastewater, reducing pollutant content and meeting discharge standards. The aeration unit is the main site for biochemical reactions such as COD degradation, nitrification, and phosphorus overabsorption by polyphosphate-accumulating bacteria, making it the core process for removing key pollutants like carbon, nitrogen, and phosphorus from wastewater treatment plants. However, traditional aeration processes account for 30% to 70% of the total energy consumption of wastewater treatment plants, and the control of aeration volume in traditional aeration modes relies heavily on manual experience. To ensure effluent quality meets standards, excessive and constant aeration is prevalent, with some even maintaining full-power blowers, resulting in significant energy waste. It is estimated that billions of kilowatt-hours of electricity are wasted annually in the aeration stage alone. Excessive aeration not only wastes energy, but excessively high dissolved oxygen (DO) levels can also damage the denitrification environment, cause sludge aging and floating, and increase the dosage of chemicals in subsequent flocculation and sedimentation processes. Therefore, the refined control of the aeration process is a key development direction for the high-quality development and intelligent transformation of wastewater treatment plants, and an inevitable requirement for the green and low-carbon development of the wastewater sector under the "dual carbon" background.

[0003] However, wastewater treatment is a complex nonlinear system characterized by large time-varying variations, strong coupling, and high disturbance. The aeration process is affected by numerous factors, including influent flow rate, influent composition, pollutant concentration, microbial concentration, and microbial activity. The pollutant concentration, required oxygen, DO consumption rate, and even microbial concentration (affected by sludge discharge and sludge return rate) within the aeration unit are all dynamically changing, making it very difficult to calculate the accurate DO requirement. Furthermore, wastewater treatment plants typically only install relatively complete instruments for measuring water quality indicators such as ammonia nitrogen, COD, total nitrogen, and total phosphorus at the influent and effluent outlets, while these key water quality indicators are not measured before, during, or after the aeration tank. This further complicates the simulation and prediction of DO variation patterns within the aeration tank. In most cases, only DO and aeration rate (Q) are data that most wastewater treatment plants measure and that are relatively accurate and abundant.

[0004] Currently, research on precision aeration mainly focuses on two directions: mechanistic models and data models. Mechanistic models, represented by ASM (Anaerobic Scale-Based Model), have significant advantages in revealing the mechanisms of microbial biochemical reactions. Their model parameters have clear physical meanings, which helps in a deeper understanding of the treatment process. However, these models face challenges in practical applications: Firstly, the modeling process is complex, and mechanistic models generally have numerous parameters (e.g., ASM 2D includes 21 biochemical reactions, 19 components, 22 stoichiometric coefficients, and 42 kinetic parameters), making the calibration of these parameters difficult and inaccurate. Secondly, once the model is calibrated and put into use, it cannot provide effective simulations if the engineering conditions (such as biological community structure, biochemical reaction kinetic parameters, etc.) change significantly. Therefore, the model must be calibrated frequently to dynamically meet the simulation accuracy requirements. These two factors make it difficult for ASM-type mechanistic models to be truly implemented in precision aeration.

[0005] Data models primarily include artificial intelligence models based on deep neural networks, which learn, find, and abstract features from massive amounts of data, thereby forming a powerful descriptive ability. In contrast, data-driven models can directly capture the nonlinear relationship between model input and output without requiring a deep understanding of the underlying biochemical reaction mechanisms. This type of model has significant advantages in handling complex systems and nonlinear relationships, and its dependence on data is relatively low. However, data-driven models also face some challenges, such as model generalization ability, data quality, and model interpretability. Currently, wastewater treatment plants generally lack data monitoring, with a severe shortage of data types in aeration tanks, a problem difficult to solve in engineering. This is because, on the one hand, due to the cost of measuring instruments and maintenance, it is difficult to obtain large amounts of data from engineering projects; on the other hand, the aeration process involves multiple complex mass transfer-reaction coupling processes and other biochemical reactions, with numerous related influencing factors. Using only the available DO and aeration rate Q, the severe lack of data types makes it difficult to achieve acceptable simulation results.

[0006] Considering current research progress, mechanistic models, based on the underlying principles of microbial biochemical reaction mechanisms and utilizing mature mechanistic knowledge in the wastewater field, suffer from complex parameters and are difficult to update. Data models, on the other hand, can automatically calculate suitable model parameters using machine learning and other methods, but they require high-quality data, and existing data types are insufficient to support accurate model construction. The synergy between data and knowledge is a current hot topic and trend in wastewater treatment process modeling research. Many studies rely on purely data-driven methods such as neural networks, which, while advantageous in handling complex nonlinear relationships, often neglect the knowledge of domain experts. Complex phenomena in wastewater treatment processes, such as oxygen dissolution and transport, and microbial metabolism, all exhibit certain biochemical reaction patterns, but these patterns have not been effectively integrated into the models, leading to poor model performance under complex environmental changes.

[0007] In summary, while the field of wastewater treatment possesses a wealth of knowledge on biochemical reactions, such as oxygen mass transfer-reaction coupling models and dissolved oxygen variation patterns, this expertise has not been fully integrated with data-driven models. For instance, features such as the oxygen mass transfer coefficient (KLa), DO change rate, and oxygen consumption rate have a significant impact on the aeration process, but data models alone cannot accurately uncover and effectively utilize these valuable characteristics. However, by using mechanistic knowledge for analysis and guidance to selectively mine these strongly correlated, high-value data features and then inputting them into the data model for training, the model's effectiveness can be further improved. Summary of the Invention

[0008] The present invention aims to provide a knowledge-assisted method for accurate aeration modeling of urban wastewater treatment, in order to solve the problems of complex parameters, difficulty in updating, and difficulty in accurately constructing data models with engineering data in existing methods.

[0009] The knowledge-assisted precise aeration modeling method for urban wastewater treatment in this scheme includes the following steps: Step 1: Establish a data-driven model based on the key data of the aeration process obtained; Step 2: Based on the mechanism model of oxygen mass transfer-reaction coupling, guide the targeted mining of key data in the aeration process, extract knowledge, and calculate new characteristic parameters related to the aeration process, including dissolved oxygen concentration change rate, dissolved oxygen consumption, oxygen consumption rate, and the rate of change of dissolved oxygen concentration change rate, forming a multi-dimensional feature matrix. Step 3: The newly added multidimensional feature values ​​are used together with the original key data of the aeration process for training and optimization of the data-driven model to obtain a knowledge-assisted aeration model.

[0010] The beneficial effects of this plan are: By leveraging mechanistic knowledge, richer data features during the aeration process can be extracted in a targeted manner, thereby improving modeling effectiveness. Deep integration of knowledge and data, with knowledge-guided extraction of in-depth data features, enables better simulation control even under conditions of widespread data loss.

[0011] Furthermore, in step 1, measured data from the wastewater treatment plant are collected at set intervals within a preset time period. The preset time period and set intervals are set according to the standard of collecting sufficient data volume.

[0012] The beneficial effect is that by designing the duration of data acquisition, a sufficient amount of data can be collected within a relatively short target time period.

[0013] Furthermore, in step 1, the key data of the aeration process include the dissolved oxygen concentration and aeration rate obtained from monitoring the aeration tank.

[0014] The beneficial effect is that the selection of key data in the aeration process can ensure the quantity, accuracy, and representativeness of the basic data.

[0015] Furthermore, in step 1, the data-driven module includes an LSTM time-series data model.

[0016] The beneficial effect is that by setting up this model, the temporal characteristics and long-term dependencies between data during the aeration process can be accurately captured.

[0017] Furthermore, in step 2, the oxygen mass transfer-reaction coupling model is expressed as: ; Where dC / dt is the rate of change of dissolved oxygen concentration over time, C is the current dissolved oxygen concentration, OUR is the oxygen consumption rate, and KLa is the oxygen mass transfer coefficient. It is the saturated dissolved oxygen concentration.

[0018] The beneficial effect is that the oxygen mass transfer-reaction coupling model is used to describe the oxygen transport process, especially the contact efficiency between oxygen and water during aeration. It can accurately describe the corresponding relationship and fully extract the required parameters.

[0019] Furthermore, in step 3, the mechanism knowledge-assisted process is as follows: guided by the oxygen mass transfer-reaction coupling model, several new features closely related to the aeration process are targeted for mining and calculation, including the rate of change of dissolved oxygen concentration, dissolved oxygen consumption, oxygen consumption rate, and the rate of change of dissolved oxygen concentration. These newly generated features are used together with the original data for training and optimization of the data model, thereby improving the performance of the data model.

[0020] The beneficial effects are: by incorporating feature extraction aided by mechanistic knowledge, we can target and deeply mine the hidden patterns and potential features behind the data, and more effectively and accurately reflect the state information such as mass transfer, accumulation and consumption of dissolved oxygen during the aeration process, thereby helping to enhance the model's ability to capture nonlinear and time-varying features and improve the model's performance. Attached Figure Description

[0021] Figure 1 This is a flowchart of an embodiment of the knowledge-assisted precise aeration modeling method for urban wastewater treatment according to the present invention; Figure 2 This is a diagram illustrating the aeration process mechanism in an embodiment of the knowledge-assisted precise aeration modeling method for urban wastewater treatment according to the present invention. Figure 3 This is a schematic block diagram illustrating the knowledge-assisted modeling in an embodiment of the knowledge-assisted precise aeration modeling method for urban wastewater treatment according to the present invention. Detailed Implementation

[0022] The following detailed description provides further details on specific implementation methods.

[0023] A knowledge-assisted method for precise aeration modeling in urban wastewater treatment, such as Figure 1 As shown, it includes the following steps: Step 1: Establish a data-driven model based on the acquired key data of the aeration process. The data-driven model is an LSTM model of the existing time-series data model. The LSTM model captures the long-term dependencies between data during the aeration process. Measured data from the wastewater treatment plant is collected at set intervals within a preset time period. The preset time period and set interval are set to ensure sufficient data volume, for example, data is collected every 2 minutes within a 24-hour preset time period. Key data of the aeration process include dissolved oxygen concentration (DO) and aeration rate (Q) monitored from the aeration tank.

[0024] The prediction accuracy of the LSTM model was significantly improved by introducing multiple internal features and external parameters. In particular, the introduction of internal features such as reaction amount and dissolved oxygen change rate effectively optimized the model's predictive power. Reaction amount is an important indicator of microbial metabolic activity during wastewater treatment, directly affecting the oxygen consumption rate and thus the changes in dissolved oxygen. Introducing this feature allows the LSTM model to better capture the dynamic changes in dissolved oxygen concentration, especially the fluctuations caused by biodegradation activities during wastewater treatment.

[0025] The dissolved oxygen change rate reflects the speed at which dissolved oxygen concentration changes over time, helping LSTM models identify patterns of dissolved oxygen concentration fluctuations. Especially during aeration, rapid changes in oxygen concentration often impact the model. By introducing this feature, the model can respond more sensitively to instantaneous changes in dissolved oxygen, further improving the prediction accuracy for sudden events such as sludge discharge and flow fluctuations. By incorporating these key internal features, LSTM models can significantly improve the accuracy and stability of predictions when processing data from wastewater aeration processes.

[0026] Step 2: Based on the oxygen mass transfer-reaction coupling mechanism model, guide the targeted mining of key data in the aeration process, extract knowledge, and calculate new characteristic parameters related to the aeration process, including the rate of change of dissolved oxygen concentration, dissolved oxygen consumption, oxygen consumption rate, and the rate of change of the rate of change of dissolved oxygen concentration (i.e., the second derivative of dissolved oxygen concentration), forming a multi-dimensional feature matrix. For example... Figure 2 As shown, the oxygen mass transfer-reaction coupling model is expressed as follows: ; Where dC / dt is the rate of change of dissolved oxygen concentration over time, C is the current dissolved oxygen concentration, OUR is the oxygen consumption rate, and KLa is the oxygen mass transfer coefficient. It is the saturated dissolved oxygen concentration.

[0027] Factors such as the oxygen mass transfer coefficient (KLa), DO change rate, and oxygen consumption rate affect aeration efficiency during wastewater treatment. The oxygen mass transfer-reaction coupling model provides LSTM models with biochemical reaction knowledge, helping to improve prediction accuracy when handling complex data.

[0028] The oxygen mass transfer-reaction coupling model describes how oxygen enters and is consumed in water through aeration. Specifically, the model consists of two parts: Oxygen transfer process: The oxygen transfer rate describes the efficiency with which oxygen is transported from the air to water. It represents the difference between the dissolved oxygen concentration in the water and its saturation concentration; the greater the difference, the faster the oxygen dissolves from the air into the water.

[0029] Oxygen consumption process: (Oxygen consumption rate) represents the amount of oxygen consumed by microbial metabolic activities and other processes. In wastewater treatment, microbial activities consume oxygen, leading to a decrease in dissolved oxygen concentration.

[0030] This means that the dissolved oxygen introduced during aeration has two main paths: part of it enters the solution, causing a change in the dissolved oxygen concentration, and the other part is not absorbed by the solution and diffuses into the air.

[0031] The amount of oxygen introduced into the solution is equal to the sum of two parts: the amount of dissolved oxygen consumed by microorganisms during metabolism (reaction amount), and the change in dissolved oxygen concentration (cumulative amount).

[0032] The process aided by mechanistic knowledge involves: guided by the oxygen mass transfer-reaction coupling model, targeted mining and calculation of several new features closely related to the aeration process, including the rate of change of dissolved oxygen concentration, dissolved oxygen consumption, oxygen consumption rate, and the rate of change of dissolved oxygen concentration (the second derivative of dissolved oxygen concentration). These newly generated features are then used together with the original data for training and optimization of the data model, thereby improving its performance. These features are obtained through targeted mining guided by knowledge, and cannot be obtained by the data-driven model alone.

[0033] The oxygen mass transfer-reaction coupling model provides essential biochemical support for LSTM, enabling it to more accurately predict changes in aeration rate when processing data from wastewater aeration processes. While LSTM can identify nonlinear characteristics and time-series relationships by analyzing historical data, the oxygen mass transfer-reaction coupling model helps improve its predictive capabilities for complex data by providing the physical context of oxygen transfer and dissolved oxygen changes.

[0034] Step 3, as follows Figure 3 As shown, the newly added multidimensional feature values ​​are used together with the original key data of the aeration process for training and optimization of the data-driven model, resulting in a knowledge-assisted aeration model.

[0035] Guided by the aforementioned oxygen mass transfer-reaction coupling process model, novel features such as the rate of change of dissolved oxygen concentration, dissolved oxygen consumption, oxygen consumption rate, and the rate of change of dissolved oxygen concentration (the second derivative of dissolved oxygen concentration) were specifically mined and calculated. These features effectively reflect the dynamic changes of dissolved oxygen and oxygen transport efficiency during aeration, improving the model's ability to capture nonlinear and time-varying characteristics. By introducing these specifically extracted features, the LSTM model showed a significant improvement in prediction accuracy during training, especially in handling dynamic changes and complex environments, where the prediction results were more accurate and stable.

[0036] The oxygen mass transfer-reaction coupling model describes the oxygen mass transfer reaction process in an aeration tank and is used to guide and predict the aeration process in the aeration tank. However, this application requires the measurement of a series of parameters to achieve the desired result. The method in this embodiment does not require the measurement of these parameters. It can discover new features by simply using the relationship provided by the oxygen mass transfer-reaction coupling model. The model finds a suitable set of parameters during the training process and establishes an accurate mapping relationship, which greatly reduces the difficulty of parameter measurement and improves the accuracy of parameter measurement.

[0037] In the example, the aeration-related data of the target wastewater treatment plant only had two columns: dissolved oxygen concentration (DO) and aeration rate (Q). Guided by the analysis of the oxygen mass transfer-reaction coupling model, several new data types were added, as shown in Table 1. The prediction results reflect the effectiveness of the model. In terms of model evaluation, the initial model's MSE was 14787, indicating a large prediction error; the RMSE was 121, indicating a large difference between the predicted and actual values; and the MAE was 87, indicating an average prediction error of 87. A value of 0.89 indicates a high degree of model fit, but there is still room for optimization. Although the model has learned the main trends in the data, it may exhibit significant errors with limited data, affecting prediction accuracy.

[0038] Table 1. Data Types Added After Knowledge Assistance

[0039] The initial measured data are DO and Q. dc / dt represents the first derivative of DO, corresponding to the left side of the equation. The reaction amount is the difference between each measurement of DO. dc / dt / dt represents the second derivative of DO. From a kinetic perspective, a new characteristic column is obtained. C in the formula corresponds to the measured data DO, and C* indicates that the saturated dissolved oxygen content is a fixed value because it only changes with temperature. The saturated dissolved oxygen concentration at 21 degrees Celsius is used, and KLa is assumed to be 0.02. At this step, all the parameters in the formula have been obtained. The oxygen consumption rate OUR is then finally calculated using the formula. The two initial data type parameters have been expanded to six columns of data, and each expanded column improves the accuracy of the data model, demonstrating the effective guidance of knowledge-based models.

[0040] By analyzing the main processes and mechanisms of aeration, multidimensional feature values ​​can be obtained. These newly added multidimensional feature values ​​are used to assist in modeling the data model. Compared with the initial model, the accuracy of the model is significantly improved, indicating that knowledge assistance has a positive impact on the establishment of the data model.

[0041] The data-driven model was compared with the oxygen mass transfer model, and the results are shown in Table 2.

[0042] Table 2 Comparison of Model Accuracy After Knowledge Assistance

[0043] Table 2 shows that the error indicators continued to decrease. The MAE (Mean Absolute Error) decreased, indicating a significant reduction in the average deviation between the model's predicted and actual values. The MSE / RMSE (Mean Squared Error / Root Mean Squared Error) ratio also decreased. The sensitivity to outliers decreased significantly, with a large decrease in MSE. Explanatory power indicators continued to improve. The coefficient of determination has increased; the model's ability to explain changes in the data has improved.

[0044] After adding the rate of change (acceleration) of dissolved oxygen, all error indices decreased. This indicates that the dynamic changes in dissolved oxygen provide additional and valuable information for predicting aeration rates, and the model captures more refined dynamic characteristics. The addition of OUR (Oxygen Requirement) is a key turning point in model optimization. This shows that the actual oxygen consumption rate of microorganisms is an extremely important driving factor for predicting the required aeration rate. It is directly related to the system's biological activity and oxygen demand, providing one of the most crucial input information for the model. As more physically or biologically meaningful features are gradually added, the model's predictive ability shows a clear and significant improvement trend. This verifies that the feature engineering approach is correct and efficient.

[0045] LSTM models make predictions based on historical data, while oxygen mass transfer-reaction coupling models provide LSTM with the physical basis of variables such as dissolved oxygen concentration. Combining and optimizing these two models can enhance their accuracy and robustness.

[0046] Compared with existing technologies, this embodiment analyzes the oxygen mass transfer-reaction coupling model in the aeration tank of a typical activated sludge wastewater treatment system, and specifically mines more data types related to aeration (including but not limited to oxygen consumption rate, dissolved oxygen change rate, change lag time, etc.). This compensates for the lack of data types and the resulting poor model accuracy, and assists in guiding the establishment of the data model. The data model can more deeply describe the aeration and reaction process in the system, thereby establishing an intelligent method for precise aeration in wastewater treatment. Through precise aeration, the wastewater treatment plant can achieve the goal of energy saving and consumption reduction while ensuring that the effluent quality meets the standards.

[0047] Compared with existing ASM modeling methods, the method in this embodiment requires less work and has a lower technical threshold in model building and calibration. Furthermore, this method uses only a few types of data—aeration rate and dissolved oxygen concentration—eliminating the need for additional measurement data during operation. These two types of data are very common and readily available, requiring no additional measurement operations and meeting the needs of most wastewater treatment plants, thus improving the method's operability, reducing operational complexity, and broadening its applicability. The method in this embodiment is based on an oxygen mass transfer-reaction coupling model, facilitating the calculation of new features and significantly reducing the computational load. This method addresses the dynamic evolution characteristics of wastewater treatment systems in actual operation; the model can dynamically evolve according to actual needs to accurately address model drift issues. That is, as new data is collected, the model gradually shifts from previous operating conditions to more recent conditions, improving the accuracy of the calculation results. The method in this embodiment considers a commonly used scenario where data types are severely lacking. It overcomes the difficulties in modeling and optimization, as well as the problems of unsatisfactory model and optimization results, when only aeration rate and dissolved oxygen concentration are known, and the influencing factors behind aeration rate and dissolved oxygen concentration, such as sludge concentration, microbial activity, and pollutant concentration, are unknown. Ultimately, it guides and assists in the establishment and optimization of the model in the most common scenario, with the fewest types of data, and without increasing the amount of measurement data.

[0048] The above descriptions are merely embodiments of the present invention, and common knowledge regarding specific structures and characteristics is not elaborated upon here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the structure of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A knowledge-based auxiliary urban sewage treatment precise aeration modeling method, characterized in that, The method comprises the following steps: Step 1, establishing a data-driven model according to the obtained key data of the aeration process; Step 2, based on the mechanism model of oxygen mass transfer-reaction coupling, guiding the directional mining of the key data of the aeration process, performing knowledge extraction, and calculating new characteristic parameters associated with the aeration process, including the change rate of dissolved oxygen concentration, the dissolved oxygen consumption, the oxygen consumption rate, and the change rate of the change rate of the dissolved oxygen concentration, to form a multi-dimensional feature matrix; Step 3, using the newly added multi-dimensional feature values and the original key data of the aeration process together for training and optimization of the data-driven model, to obtain a knowledge-assisted aeration model.

2. The knowledge-based urban wastewater treatment precise aeration modeling method according to claim 1, characterized in that: In step 1, the measured data of the sewage treatment plant is collected at a set interval within a preset time length, and the preset time length and the set interval are set according to the standard of collecting sufficient data volume.

3. The knowledge-based urban wastewater treatment precise aeration modeling method according to claim 2, characterized in that: In step 1, the key data of the aeration process includes the dissolved oxygen concentration and the aeration quantity monitored from the aeration tank.

4. The knowledge-based urban wastewater treatment precise aeration modeling method according to claim 2, characterized in that: In step 1, the data-driven model includes an LSTM time series data model.

5. The knowledge-based urban wastewater treatment precise aeration modeling method according to claim 1, characterized in that: In step 2, the oxygen mass transfer-reaction coupling process follows the double membrane theory, describes the complex process of oxygen from the inside of the aeration micro-bubble passing through the gas-liquid interface into the liquid phase and finally being consumed by microorganisms, and the process of mass transfer and reaction coupling can be described by the following basic equation: ； where dC / dt is the rate of change of dissolved oxygen concentration with time, C is the current dissolved oxygen concentration, OUR is the oxygen uptake rate, KLa is the oxygen mass transfer coefficient, is the saturated dissolved oxygen concentration.

6. The knowledge-based urban wastewater treatment precise aeration modeling method according to claim 1, characterized in that: In step 3, the mechanism knowledge assistance process is as follows: according to the guidance of the oxygen mass transfer-reaction coupling model, a plurality of new features closely related to the aeration process are calculated, including the change rate of dissolved oxygen concentration, the dissolved oxygen consumption, the oxygen consumption rate, and the change rate of the change rate of the dissolved oxygen concentration, and these newly generated features are used together with the original data for training and optimization of the data model.

Images