Multi-source data fusion modeling method and system in aeration process

Abstract

The invention provides a multi-source data fusion modeling method and system in an aeration process, and is applied to the field of intelligent aeration control in sewage treatment. The method comprises the steps that multi-source time sequence data such as dissolved oxygen, turbidity, flow, temperature, power and pool bottom pressure pulsation signals are collected, dissolved oxygen response lag is calculated through cross-correlation analysis with power change as the reference, time sequence alignment is carried out, and a dissolved oxygen reference interval is predicted by utilizing calibration data in combination with a physical constraint LSTM model; performing spectral analysis on the pressure pulsation signal to extract a gas-liquid coupling characteristic value, and generating a cooperative regulation instruction of the frequency of the blower and the rotating speed of the stirrer based on the information; by means of the scheme, control oscillation caused by lag of the dissolved oxygen sensor can be effectively overcome, online monitoring of bubble form distribution is achieved, the gas-liquid mass transfer efficiency is improved, invalid aeration is avoided, and system energy consumption is remarkably reduced on the premise that stable effluent quality is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, and in particular to a multi-source data fusion modeling method and system for aeration processes. Background Technology

[0002] The efficiency of biochemical reactions in wastewater treatment depends on the full contact between activated sludge and dissolved oxygen. Among them, the aeration system is the most energy-consuming unit, accounting for 50% to 60% of the total energy consumption of the plant.

[0003] Existing control technologies mostly employ proportional-integral-derivative (PID) control or feedback regulation based on dissolved oxygen setpoints, maintaining stable dissolved oxygen concentrations by adjusting blower frequency. However, these methods have significant drawbacks. Dissolved oxygen sensors exhibit a physical response lag of three to ten minutes to changes in oxygen levels in the water, while blower power regulation is instantaneous. This leads to a time asynchrony between the input signal and output response of the control system, easily causing overshoot and continuous oscillations. Furthermore, traditional monitoring methods can only obtain air supply and residual dissolved oxygen values, failing to perceive the actual distribution morphology of bubbles in the turbid liquid phase. When the aeration device becomes clogged or damaged, producing inefficient large bubbles, the system still blindly increases the air volume, resulting in energy waste. Simultaneously, relying solely on dissolved oxygen data makes it difficult to distinguish between dissolved oxygen fluctuations caused by influent load fluctuations and those caused by changes in ambient temperature, leading to misjudgments and inappropriate regulation. While existing data-driven models based on artificial neural networks possess nonlinear fitting capabilities, they lack physical constraints and may output predicted values ​​exceeding the saturated dissolved oxygen concentration, violating Henry's Law and causing control commands to fail. These problems collectively constrain the energy efficiency optimization and operational stability of the aeration process.

[0004] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a multi-source data fusion modeling method and system for the aeration process, aiming to solve the problems in the prior art, such as control oscillation caused by the response lag of dissolved oxygen sensors, the inability to monitor the bubble morphology in turbid wastewater leading to unknown gas-liquid mass transfer efficiency, the difficulty of distinguishing complex working conditions by a single data source, and the tendency of pure data-driven models to output predicted values ​​that violate physical laws. By introducing time-series alignment correction, spectral feature extraction based on high-frequency pressure pulsation signals, neural network modeling based on physical constraints, and a dual-channel collaborative control strategy, the present invention achieves accurate prediction and stable control of dissolved oxygen concentration during the aeration process, and effectively identifies the bubble distribution state without relying on visual perception, thereby improving oxygen mass transfer efficiency and reducing system energy consumption.

[0006] This invention provides a multi-source data fusion modeling method for an aeration process, which is applied to the aeration stage of a wastewater treatment system and includes the following steps: Multi-source time series data were collected during the wastewater treatment process. The multi-source time series data included dissolved oxygen concentration, water turbidity, influent flow rate, ambient temperature, operating power of aeration equipment, and liquid pressure pulsation signal at the bottom of the tank. The change sequence of operating power is selected as a reference benchmark. The time lag parameter of dissolved oxygen concentration relative to operating power is calculated. The time lag parameter is used to perform time-series alignment correction on the multi-source time series data to obtain the calibrated multi-source time series data. A neural network model for predicting oxygen demand was constructed. The water turbidity, influent flow rate and ambient temperature in the calibrated multi-source time series data were used as input features to input the neural network model and output the baseline range of dissolved oxygen concentration under the current operating conditions. Spectral features were extracted from the liquid pressure pulsation signal at the bottom of the pool to obtain gas-liquid coupling feature values ​​that characterize the morphological distribution of bubbles in wastewater. Based on the dissolved oxygen concentration benchmark range, real-time collected dissolved oxygen concentration, and gas-liquid coupling characteristic values, operating frequency adjustment commands for aeration equipment and start / stop control commands or speed adjustment commands for auxiliary stirring equipment are generated to maintain dissolved oxygen stability during wastewater treatment.

[0007] In some optional embodiments, calculating the time lag parameter of dissolved oxygen concentration relative to operating power includes: Extract the operating power sequence and dissolved oxygen concentration sequence within a historical time window of a preset length; Calculate the cross-correlation function between the operating power sequence and the dissolved oxygen concentration sequence; Search for the peak value of the cross-correlation function and determine the time offset corresponding to the peak value as the time lag parameter; By shifting the timestamps of the dissolved oxygen concentration sequence forward by the time lag parameter, a calibration dataset for physical response synchronization is established.

[0008] In some optional embodiments, after obtaining the calibrated multi-source time series data, the method further includes: A state-space equation is constructed, with the actual dissolved oxygen concentration set as the system state variable and the sensor measurement noise set as the observation noise. The extended Kalman filter algorithm is used to iteratively estimate the dissolved oxygen concentration data in the calibrated multi-source time series data, and high-frequency random noise is filtered out while preserving the signal phase characteristics.

[0009] In some optional embodiments, the neural network model employs a long short-term memory network architecture, and the construction process of the neural network model includes: Extract the water turbidity change rate and influent flow rate change rate from historical operational data as dynamic feature vectors; The dissolved oxygen concentration at historical moments is used as an autoregressive feature vector; A long short-term memory network was trained using historical operational data. The mean square error between the predicted output and the actual dissolved oxygen concentration was minimized using the backpropagation algorithm, and the network weight parameters were established.

[0010] In some optional embodiments, the training process of the neural network model further includes a physical constraint correction step: The theoretical saturated dissolved oxygen concentration was calculated using Henry's Law based on real-time ambient temperature data. A penalty term is added to the loss function of the neural network model. When the predicted value output by the neural network model exceeds the theoretical saturated dissolved oxygen concentration, the loss function value is increased to limit the output of the neural network model to conform to physical laws.

[0011] In some optional embodiments, spectral feature extraction of the pool bottom liquid pressure pulsation signal includes: A fast Fourier transform is performed on the pressure pulsation signal at the bottom of the pool to convert the time-domain signal into a frequency-domain power spectrum. The ratio of the energy value in the preset high-frequency range to the total energy value across the entire frequency band is calculated by integral calculation. The ratio is determined as a gas-liquid coupling characteristic value to characterize the uniformity of bubble size generated by aeration.

[0012] In some optional embodiments, generating operating frequency adjustment commands for the aeration equipment and start / stop control commands or speed adjustment commands for the auxiliary mixing equipment includes: When the gas-liquid coupling characteristic value is lower than the preset uniformity threshold, the aeration state is determined to be bubble aggregation or uneven distribution. During aeration, if the real-time dissolved oxygen concentration is lower than the lower limit of the dissolved oxygen concentration reference range, an instruction is generated to maintain or reduce the operating frequency of the aeration equipment, and an instruction is generated in parallel to start or increase the speed of the auxiliary stirring equipment, using the shear force of the water flow to break up the polymerized bubbles.

[0013] In some optional embodiments, the method further includes a condition attribution step based on correlation analysis: When the real-time dissolved oxygen concentration is lower than the dissolved oxygen concentration benchmark range, calculate the slope of change in ambient temperature and the slope of change in operating power. If the slope of the change in ambient temperature is positive and the slope of the change in operating power is not positive, it is determined that the oxygen mass transfer coefficient is reduced due to the environmental thermal effect, and a temperature compensation command to increase the operating frequency of the aeration equipment is generated.

[0014] In some optional embodiments, the condition attribution step further includes: If the slope of change in ambient temperature is within the preset range, but the slope of change in water turbidity exceeds the abrupt change threshold, it is determined to be an influent organic load shock. Generate feedforward control commands to increase the operating frequency of the aeration equipment in advance, before the dissolved oxygen concentration actually decreases.

[0015] In some optional embodiments, the method further includes a correction step for rainfall disturbances: Obtain real-time rainfall data from external meteorological data; When the real-time rainfall exceeds the preset value and the inflow rate increases while the water turbidity decreases, it is determined to be a rainwater dilution condition. The output of the neural network model is corrected to lower the lower limit of the dissolved oxygen concentration baseline range, thus avoiding excessive operation of the aeration equipment and resulting energy waste.

[0016] In some optional embodiments, generating operating frequency adjustment commands for the aeration equipment includes optimization steps based on energy consumption ratio: The energy consumption required to remove a unit of pollutant per unit time is calculated as the real-time energy efficiency ratio. A mapping relationship between the operating frequency adjustment and the real-time energy efficiency ratio change was established using a support vector machine regression algorithm. Under the premise of meeting the dissolved oxygen concentration baseline range, the operating frequency that minimizes the real-time energy efficiency ratio is searched as the target control value.

[0017] In some optional embodiments, the method further includes a device health status monitoring step: Collect mechanical vibration signals during the operation of the aeration equipment; Extract the amplitude of the mechanical vibration signal at its fundamental frequency and its harmonics; When the amplitude exceeds the health threshold, the rate of increase of frequency adjustment is limited when generating the operating frequency adjustment command to prevent further mechanical damage to the equipment.

[0018] In some optional embodiments, the method further includes a time-segmented control step based on K-means clustering: Extract the daily variation characteristics of historical operation data, and use the K-means clustering algorithm to divide the entire day's operation time into three time periods: low load, medium load, and high load. For each time period category, different confidence thresholds for the neural network model are set. In the low-load period category, the confidence threshold is lowered to reduce the frequency of control actions; in the high-load period category, the confidence threshold is increased to enhance the control response speed.

[0019] In some optional embodiments, the method further includes a linkage control step for the mud-water mixing state: Calculate sludge load parameters based on water turbidity and influent flow rate; When the sludge load parameters indicate that the influent concentration is too high, a command is generated to adjust the flow rate of the sludge return pump, and in conjunction with the operation frequency adjustment of the aeration equipment, to maintain the sludge concentration balance within the wastewater treatment system.

[0020] In some optional embodiments, the method further includes an online model update step: The multi-source time series data generated from daily actual operation and the corresponding control effects are stored in the historical database. The neural network model is incrementally trained using new data at a preset cycle, and the network weight parameters are updated to adapt to the seasonal changes in the wastewater treatment system.

[0021] This invention provides a multi-source data fusion modeling system for an aeration process, the system comprising: The multidimensional sensing interface module is configured to communicate with a group of field sensors distributed in the wastewater treatment tank. The field sensor group includes a dissolved oxygen sensor, a turbidity meter, a flow meter, a temperature sensor, a power meter, and a high-frequency pressure sensor installed at the bottom of the tank. The multidimensional sensing interface module is used to receive multi-source time series data in real time. The timing synchronization processor is configured to cache multi-source time series data and execute a cross-correlation analysis algorithm to determine the response lag time of dissolved oxygen concentration relative to operating power. Based on the response lag time, the multi-source time series data is resampled and aligned, and a physically synchronized calibration dataset is output. The intelligent fusion computing core includes a memory and a processor. The memory contains a pre-trained neural network model. The processor is configured to: input the water turbidity, influent flow rate and ambient temperature from the calibration dataset into the neural network model, calculate the dissolved oxygen reference range under the current operating conditions; and perform frequency domain transformation on the signal from the high-frequency pressure sensor to extract gas-liquid coupling feature values ​​to characterize the physical morphology of bubbles. The dual-channel control decision module is configured to receive the output results of the intelligent fusion computing core. When the gas-liquid coupling characteristic value is abnormal, it generates start-stop control commands or speed adjustment commands for the auxiliary stirring equipment; when the real-time dissolved oxygen concentration deviates from the dissolved oxygen reference range, it generates frequency adjustment commands for the aeration equipment. The execution drive interface is connected to the frequency converter controller of the aeration equipment and the driver of the auxiliary mixing equipment, respectively, and is used to issue frequency adjustment commands and speed adjustment commands.

[0022] In some optional embodiments, the intelligent fusion computing core also integrates a physical constraint logic unit; The physical constraint logic unit is configured to store the Henry's Law calculation formula and calculate the theoretical saturated dissolved oxygen threshold based on the real-time collected ambient temperature. When the upper limit of the dissolved oxygen baseline range output by the neural network model exceeds the theoretical saturated dissolved oxygen threshold, the physical constraint logic unit triggers a signal truncation mechanism to forcibly clamp the upper limit of the dissolved oxygen baseline range to the theoretical saturated dissolved oxygen threshold, in order to prevent the control command from exceeding the physical limit.

[0023] In some optional embodiments, the dual-channel control decision module is configured with collaborative optimization logic; The collaborative optimization logic is programmed to perform the following judgment: if the gas-liquid coupling characteristic value is lower than the preset threshold, it indicates that the bubble breakage is insufficient. At this time, the gain of the frequency adjustment command sent to the aeration equipment is locked or reduced, and a signal to increase the speed of the auxiliary stirring equipment is sent in parallel through the execution drive interface until the gas-liquid coupling characteristic value is restored to the preset threshold.

[0024] In some optional embodiments, the timing synchronization processor is internally configured with an adaptive Kalman filter; The adaptive Kalman filter constructs a state equation with the actual dissolved oxygen concentration as the state variable and the sensor noise as the observation variable. It is configured to perform recursive filtering on the calibration dataset after data resampling and alignment to provide a high signal-to-noise ratio input vector to the intelligent fusion computing core.

[0025] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention.

[0026] The multi-source data fusion modeling method and system for an aeration process of the present invention has the following beneficial effects: This invention addresses the problems of lag in dissolved oxygen (DO) control, unmeasurable gas-liquid mass transfer efficiency, and model output violations of physical laws during wastewater treatment aeration by integrating multi-source data and physical constraint modeling. It utilizes time-series alignment technology to eliminate the impact of sensor response delays and extracts bubble morphology features based on pressure pulsation spectra, enabling online sensing of bubble distribution in turbid water. A Henry's Law-constrained LSTM model predicts the DO baseline range, ensuring the output conforms to physical limits. Dual-channel coordinated control of the blower and agitator improves oxygen mass transfer efficiency. This effectively reduces DO fluctuations by over 40%, reduces overall energy consumption by 15%-20%, and significantly improves system stability and energy efficiency. Attached Figure Description

[0027] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0028] Figure 1 This is a flowchart of a multi-source data fusion modeling method for an aeration process according to an embodiment of the present invention; Figure 2This is a schematic diagram of the structure of a multi-source data fusion modeling system for the aeration process according to an embodiment of the present invention. Detailed Implementation

[0029] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the invention will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0030] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0031] The flowchart shown in the attached diagram is merely an illustrative example and does not necessarily include all steps. For example, some steps may be broken down, while others may be combined or partially combined. Therefore, the actual execution order may change depending on the specific circumstances.

[0032] In the aeration process of wastewater treatment, there is a significant time lag between changes in dissolved oxygen concentration and the operating power of the blower. This is due to the physical delay caused by the slower response speed of the dissolved oxygen sensor compared to the gas delivery process. If this lag is not corrected, directly using the original synchronization timestamp data for modeling or control will result in a causal mismatch between input and output, leading to overshoot or oscillation in the control system. By identifying the maximum correlation delay between power changes and DO response through cross-correlation analysis, and aligning the DO data accordingly, the physical synchronization of the data sequences involved in the modeling can be restored, thereby improving the accuracy of model predictions and the stability of the control system. Furthermore, gas-liquid mass transfer efficiency depends not only on the gas supply but also highly on the size distribution and dispersion uniformity of bubbles. Traditional visual methods cannot effectively observe bubble morphology in high-turbidity wastewater. Based on fluid dynamics principles, the bubble bursting and movement process generates pressure pulsations of specific frequencies in the liquid. High-frequency components correspond to the shearing behavior of tiny bubbles, while low-frequency components reflect the surging phenomenon of large bubbles or gas masses. By performing spectral analysis on the high-frequency pressure signal at the bottom of the pool and extracting the energy ratio of different frequency bands, the bubble distribution state can be indirectly quantified, enabling non-invasive monitoring of the mass transfer interface characteristics. Furthermore, the limiting concentration of dissolved oxygen is constrained by temperature and gas pressure, following the thermodynamic equilibrium relationship determined by Henry's Law. Introducing this physical constraint into the data-driven model can prevent the neural network from outputting unreasonable predictions exceeding saturated dissolved oxygen through a penalty mechanism, ensuring that the model behavior conforms to natural laws. By integrating multi-source heterogeneous sensor information, aligning dynamic responses in the time dimension, fusing gas-liquid two-phase state characteristics in the spatial dimension, and combining the dynamic changing trend of biochemical oxygen demand, the control system can distinguish between different operating conditions such as true anoxic conditions, environmental disturbances, and equipment malfunctions. This allows for the implementation of differentiated dual-channel regulation strategies—adjusting both gas supply intensity and water flow shear conditions to optimize bubble breakage—thereby reducing ineffective energy consumption while ensuring treatment efficiency, and improving the operational robustness and energy utilization efficiency of the entire aeration system.

[0033] like Figure 1 As shown in the figure, this invention provides a multi-source data fusion modeling method for an aeration process, which includes the following steps: Step S100: Collect multi-source time series data during the wastewater treatment process.

[0034] In one embodiment, the system continuously collects multi-source time-series data, including dissolved oxygen concentration, water turbidity, influent flow rate, ambient temperature, aeration equipment operating power, and bottom liquid pressure pulsation signals, using a sensor array deployed on the aeration tank and influent pipes at a unified time reference. Dissolved oxygen concentration is acquired by an immersion optical or electrochemical sensor; water turbidity is measured by a turbidity meter based on transmitted or scattered light principles; influent flow rate is determined by an electromagnetic or ultrasonic flow meter; ambient temperature is monitored by a resistance temperature detector (RTD) or thermocouple sensor; operating power is read in real-time by a power meter installed at the output of the blower's frequency converter; and bottom liquid pressure pulsation signals are collected by a high-frequency response pressure sensor with a sampling frequency of at least 500 Hz, installed in a stable flow field region above the aeration heads, to capture localized hydrodynamic disturbances caused during bubble generation, release, and breakup.

[0035] In some other alternative implementations, the multi-source data may further include pH value, oxidation-reduction potential (ORP), sludge concentration (MLSS), or external meteorological information; the sensing method for pressure pulsation signals may be replaced with a micro-pressure detection device based on fiber Bragg gratings (FBG) to improve electromagnetic interference resistance.

[0036] Step S200: Select the change sequence of operating power as a reference benchmark, calculate the time lag parameter of dissolved oxygen concentration relative to operating power, and perform time-series alignment correction on the multi-source time series data based on the parameter to obtain the calibrated multi-source time series data.

[0037] In one embodiment, the system treats the operating power of the aeration equipment as the input excitation signal and the dissolved oxygen concentration as the system response signal, using statistical signal processing techniques to identify the dynamic response delay between the two. By performing cross-correlation analysis or other time delay estimation algorithms on historical operating data segments, the actual physical lag time of dissolved oxygen concentration relative to power change is determined. Based on this lag time, the original collected dataset is resampled or timestamped to ensure that the input features and target labels are synchronized in terms of physical causality, thereby constructing a calibration dataset that conforms to the dynamic characteristics of the real process.

[0038] In some other alternative implementations, the reference benchmark can be replaced by the blower airflow output signal or the air pipeline pressure signal; the timing alignment method can also adopt alternative algorithms such as dynamic time warping (DTW) or maximum likelihood estimation based on the state-space model to adapt to the identification of time delay characteristics under nonlinear or multimodal conditions.

[0039] Step S300: Construct a neural network model for predicting oxygen demand. Input the water turbidity, influent flow rate and ambient temperature from the calibrated multi-source time series data into the model and output the baseline range of dissolved oxygen concentration under the current operating conditions.

[0040] In one embodiment, the neural network model is a recurrent neural network architecture with long-term memory, capable of learning the dynamic load variation patterns of the wastewater biochemical reaction system. The input features primarily reflect the influent organic load level and environmental conditions. After training with historical operating data, the model can predict the theoretical range of dissolved oxygen required to maintain microbial activity under current water quality and hydraulic conditions. This baseline range comprehensively considers the model's predicted mean and its uncertainty boundary, serving as the target setpoint for subsequent control decisions.

[0041] In other alternative implementations, the neural network model may employ a gated recurrent unit (GRU), a Transformer structure, or an ensemble learning model such as random forests and gradient boosting trees; the input features may also be extended to historical DO trends, holiday pattern identifiers, or seasonal factor encoding to enhance the model’s adaptability to complex and unsteady conditions.

[0042] Step S400: Extract spectral features from the liquid pressure pulsation signal at the bottom of the pool to obtain gas-liquid coupling feature values ​​that characterize the morphological distribution of bubbles in wastewater.

[0043] In one embodiment, the system performs frequency domain transformation processing on the acquired pressure pulsation time-domain signal to analyze its energy distribution characteristics in different frequency ranges. Since there is a physical correlation between bubble size and its bursting frequency—that is, tiny bubbles generate high-frequency fluctuations, while large bubbles or air masses excite low-frequency oscillations—the average particle size and uniformity of the bubble swarm can be quantified by comparing the energy proportions of specific frequency bands. The extracted gas-liquid coupling characteristic values ​​reflect the quality of the gas dispersion state during aeration, indirectly characterizing the effective area of ​​the oxygen mass transfer interface.

[0044] In other alternative implementations, the feature extraction method may employ wavelet packet decomposition, Hilbert-Huang transform (HHT), or multi-channel blind source separation techniques to improve the analytical capability for transient events and non-stationary signals; the gas-liquid coupling feature value may also consist of multiple sub-indices, such as high-frequency energy ratio, dominant frequency offset, and spectral entropy value, which are jointly input to the downstream judgment logic.

[0045] Step S500: Based on the dissolved oxygen concentration benchmark range, the real-time collected dissolved oxygen concentration, and the gas-liquid coupling characteristic value, generate operating frequency adjustment commands for the aeration equipment and start / stop control commands or speed adjustment commands for the auxiliary stirring equipment to maintain dissolved oxygen stability during the wastewater treatment process.

[0046] In one embodiment, the control system compares the measured dissolved oxygen concentration with the baseline range output by the model in real time, and formulates a dual-channel coordinated control strategy based on the bubble distribution state reflected by the gas-liquid coupling characteristic value. When oxygen deficiency is detected and the bubbles are well dispersed, the blower frequency is adjusted first to increase the air supply; when oxygen deficiency is detected but bubble aggregation is severe, the blower frequency increase operation is suppressed, and the underwater agitator speed is increased instead to improve the bubble breaking effect by utilizing fluid shearing; when the dissolved oxygen is within the normal range but external conditions change, the system can dynamically adjust the control weights according to preset rules or optimization objectives to achieve a balance between energy saving and stability.

[0047] In other alternative implementations, the control command generation mechanism may be based on a fuzzy logic controller, model predictive control (MPC), or a multi-agent reinforcement learning framework; the regulation of the auxiliary stirring equipment may also be linked to the adjustment of the guide vane angle or the opening and closing of the aeration pipeline in sections, forming a more complex three-dimensional mixing optimization path.

[0048] The aforementioned technical features form a closed-loop synergistic mechanism: multi-source data acquisition provides a comprehensive state perception foundation for the entire system; time alignment ensures the physical consistency between input data and response results, avoiding misjudgments caused by sensor response delays; the neural network model establishes oxygen demand prediction capabilities based on calibration data, providing scientific control targets; the spectral analysis of pressure pulsation signals overcomes the limitations of traditional visual monitoring methods in turbid environments, achieving a "non-invasive" assessment of gas-liquid mass transfer efficiency; finally, the dual-channel control strategy based on multi-information fusion can distinguish between two fundamentally different causes of hypoxia—"insufficient gas supply" and "mass transfer obstruction"—and take differentiated intervention measures, significantly improving control accuracy and energy efficiency. This fully coupled design from perception and analysis to decision-making effectively solves the technical problems of response lag, black-box operation, and excessive energy consumption in existing aeration control systems.

[0049] Through the above-described scheme, this embodiment achieves high-precision, adaptive control of dissolved oxygen concentration during wastewater treatment. While ensuring stable effluent quality compliance, it significantly reduces energy waste caused by ineffective aeration and improves the operational reliability and intelligence of the aeration system. Compared with existing technologies, this embodiment eliminates control oscillations caused by sensor lag, improving the system's dynamic response performance; it achieves online monitoring of bubble morphology without relying on image recognition, enhancing diagnostic capabilities under complex operating conditions; and by introducing physical process cognition and multivariate collaborative control, it avoids the physical paradoxes that may occur in purely data-driven models, improving the safety and robustness of the overall control strategy.

[0050] In one specific implementation, the operating power sequence and dissolved oxygen concentration sequence within a historical time window of a preset length are selected as the analysis objects, including: selecting continuous... A time series segment consisting of 10 sampling points, where The value is set to 3600, corresponding to one hour of operational data, to ensure the stability of the statistical analysis. Specifically, the real-time operating power of the blower collected at 1-second intervals within this time period is recorded as a sequence. The dissolved oxygen concentration recorded synchronously is denoted as the sequence. Then, calculate the cross-correlation function between the two sequences. Its definition is as follows:

[0051] in Indicates the time offset. The maximum search delay steps are set to a value ranging from 0 to 600 seconds (i.e., 10 minutes) to cover a reasonable range of DO response lags in typical wastewater treatment plants. The search then... Offset to reach the global maximum value The time delay corresponding to this peak value is determined as the actual response lag. ,in This is the system sampling period.

[0052] Subsequently, the raw dissolved oxygen data was timestamped: the current time was corrected. Input feature vector and future Dissolved oxygen value after seconds Pairing is used to construct a calibration dataset that is physically causally consistent. For example, if the calculated... Then use during modeling Prediction of variables such as power and flow rate at any given time The actual DO value in seconds is used to eliminate input / output mismatch caused by sensor response lag.

[0053] In other alternative implementations, the cross-correlation analysis can employ a normalized cross-correlation (NCC) function to improve robustness under different operating conditions; or dynamically update the hysteresis parameters using a sliding window. It can adapt to the drift in response characteristics caused by seasonal water quality changes or equipment aging; in addition, cross-spectral analysis can be used to estimate the phase delay in the frequency domain as an alternative to time-domain cross-correlation.

[0054] Through the above scheme, this embodiment can accurately identify the dynamic response lag time of dissolved oxygen concentration relative to gas supply power, and achieve time alignment of multi-source data at the physical process level based on this parameter. Compared with the prior art, this embodiment avoids the "causal inversion" problem of the control model caused by sensor response delay, improves the training accuracy of the subsequent AI prediction model and the dynamic response performance of the control system, and requires no additional hardware investment. Time series correction can be completed solely through data analysis, demonstrating good engineering applicability and scalability.

[0055] In one specific implementation, based on the above embodiments, to address the high-frequency random noise problem contained in the dissolved oxygen concentration signal in the calibrated multi-source time series data, a nonlinear state-space model is first constructed to dynamically estimate the true dissolved oxygen concentration. Specifically, the true dissolved oxygen concentration is defined as the system state variable. Its evolution over time is described by a first-order autoregressive form:

[0056] in Assuming the process noise follows a zero-mean Gaussian distribution, the observation equation correlates sensor readings with the actual state:

[0057] in Indicates the first Dissolved oxygen concentration measurements collected by the time sensor. This is the observation matrix (simplified to 1 here). For noise measurement, zero-mean Gaussian white noise was also used. Since the system exhibits weak nonlinearity and requires real-time processing, the Extended Kalman Filter (EKF) algorithm was used to recursively solve the state-space model.

[0058] Then, the following iterative steps are performed: First, based on the state estimate from the previous time step... Covariance Perform state prediction; then calculate the Jacobian matrix to linearize the observation function and update the Kalman gain; finally, use the current observations. The prediction results are corrected, and the filtered dissolved oxygen concentration estimate is output. This process runs cyclically within each sampling period, ensuring that high-frequency random fluctuations caused by the sensor are effectively suppressed without introducing phase delay, thereby improving the signal-to-noise ratio of the data input to the subsequent neural network model.

[0059] In other alternative implementations, the state transition model can employ a second-order dynamic system to reflect the trend term of dissolved oxygen changes; or an unscented Kalman filter (UKF) can be used instead of the EKF, with Sigma point propagation handling strongly nonlinear cases more accurately; furthermore, the observation noise covariance... It can be designed as a time-varying parameter, automatically adjusted based on the historical residuals of the sensor, to achieve adaptive filtering.

[0060] Through the above scheme, this embodiment can effectively remove random noise interference in dissolved oxygen sensor data while retaining dynamic response characteristics, avoiding the time lag effect caused by traditional smoothing methods (such as moving average), thereby improving the accuracy and real-time performance of subsequent modeling and control decisions.

[0061] In one specific implementation, the neural network model employs a Long Short-Term Memory (LSTM) network architecture, and the construction process of the neural network model includes: The turbidity change rate and influent flow rate change rate are extracted from historical operational data as dynamic feature vectors. Specifically, ΔTurb is obtained by calculating the ratio of the turbidity difference between adjacent time points to the sampling period, and ΔFlow is obtained in the same way to reflect the instantaneous variation trend of the operating conditions. Then, the dissolved oxygen concentration values ​​of the previous time point and earlier are used as autoregressive feature vectors input into the model to construct the time series dependency relationship and enhance the ability to model the system inertia. Next, the time window sequence containing the above features is input into a two-layer LSTM network. The hidden layer of this network is configured with 128 neurons and uses the tanh activation function. The mean square error (MSE) between the predicted output and the actual dissolved oxygen concentration is minimized through the backpropagation algorithm, and the network weight parameters are iteratively optimized to complete the model training process.

[0062] In other alternative implementations, the dynamic feature vector may also include the rate of temperature change or the energy fluctuation rate of the pressure pulsation signal; the autoregressive feature vector can be extended to the mean or exponentially weighted moving average within a sliding window; the neural network architecture can also be replaced with a gated recurrent unit (GRU) network or a bidirectional LSTM network to adapt to wastewater treatment systems of different sizes and response characteristics.

[0063] Through the above scheme, this embodiment can effectively capture the nonlinear time-series relationship between multiple variables in the wastewater treatment process, improve the prediction accuracy of dissolved oxygen concentration change trend, and enhance the adaptability of neural network to complex operating conditions by using dynamic features and historical states for joint modeling, thus providing a reliable data foundation for subsequent control decisions.

[0064] In one specific implementation, based on the above embodiments, the training process of the neural network model further includes a physical constraint correction step to ensure that the model output conforms to basic physicochemical laws. Specifically, the theoretical saturated dissolved oxygen concentration under the current operating conditions is first calculated based on real-time collected ambient temperature data. This value is based on Henry's Law and determined through an empirical formula:

[0065] in Temperature in Celsius The measured atmospheric pressure is given in kPa. This calculation result serves as the physical upper limit boundary for the predicted output.

[0066] Then, an asymmetric penalty term is introduced during the construction of the loss function for the neural network. The traditional mean squared error loss function is extended to a composite form:

[0067] The first term minimizes the deviation between the predicted and actual values, and the second term is the physical violation penalty. When the predicted value... Exceeding the current time corresponding When the threshold is reached, the ReLU function is activated and produces a positive increment, leading to a significant increase in the total loss; conversely, if the threshold is not exceeded, the term is zero. Weighting coefficients It is set to a constant greater than zero, and its value range can be adjusted according to the actual training stability. The typical value is 10 to 100.

[0068] During the model's backpropagation optimization process, the gradient is propagated to the output and hidden layers through this loss function, forcing the network to learn the implicit rule that "DO concentration cannot exceed the saturation value," thereby avoiding the generation of invalid predictions that violate physical facts. This constraint mechanism is dynamically updated in each training iteration. Since the penalty threshold changes in real time with temperature and air pressure, it is time-varying.

[0069] In some other alternative implementations, the physical penalty term can be replaced with a quadratic penalty function. To enhance the suppression of severe exceedances, a piecewise linear weighted strategy can be adopted, so that the penalty intensity gradually increases with the extent of exceedance. In addition, the calculation of saturated dissolved oxygen can also be done without relying on empirical formulas, but by interpolating from a pre-stored temperature-dissolved oxygen reference table using a lookup method. The table data is established based on standard water chemistry handbooks or on-site calibration experiments.

[0070] Through the above approach, this embodiment can effectively prevent neural network outputs predictions that violate Henry's Law, and improve the robustness and reliability of the model under abnormal input or extrapolation scenarios. At the same time, embedding prior physical knowledge into the data-driven framework reduces the risk of control logic failure caused by pure black-box modeling, and ensures the safety boundary of the aeration control system operation.

[0071] In one specific implementation, the extraction of spectral features from the liquid pressure pulsation signal at the bottom of the aeration tank includes the following steps: First, a time-domain pressure signal is acquired by a high-frequency pressure sensor installed at the bottom of the aeration tank. This signal is continuously recorded at a sampling frequency of not less than 500 Hz to capture the fluid dynamic fluctuations caused by bubble generation, detachment, and collapse. Specifically, a Fast Fourier Transform (FFT) is performed on the time-domain signal to transform it from the time domain to the frequency domain, obtaining the corresponding power spectral density distribution, thereby revealing the energy distribution characteristics of different frequency components. Then, a preset high-frequency range is defined, such as 50 Hz to 200 Hz, which corresponds to the pressure fluctuations generated by the high-frequency oscillation and collapse of microbubbles in shear flow; the total energy integral value across the entire frequency range is considered as a normalization benchmark. Next, the ratio of the energy integral within the high-frequency range to the total energy integral across the entire frequency range is calculated, and this ratio is determined as a quantitative index characterizing the bubble size distribution state—the gas-liquid coupling characteristic value. The trend of this characteristic value reflects the uniformity and density of the bubble swarm during aeration: a high ratio indicates that high-frequency components dominate, with small and uniformly distributed bubble particles; a significantly decreased ratio suggests enhanced low-frequency energy, potentially indicating large bubble aggregation or uneven local aeration. In other optional implementations, the frequency band division can be adaptively adjusted according to the actual aeration device type and pool water depth; wavelet packet transform can be used instead of FFT to improve the local frequency energy resolution under non-stationary signals; or a bandpass filter bank combined with envelope detection can be used to achieve a similar frequency domain energy separation function.

[0072] Through the above-described scheme, this embodiment can objectively quantify the morphological distribution of bubbles during aeration based on the spectral characteristics of fluid pressure pulsation signals, overcoming the technical bottleneck of traditional visual monitoring methods failing in turbid wastewater environments. Compared with existing technologies, this embodiment can indirectly perceive the dynamic behavior of the gas-liquid mass transfer interface without relying on optical pathways or invasive probes, improving the system's adaptability and reliability under complex water quality conditions. Simultaneously, the extracted gas-liquid coupling characteristic values ​​provide process state parameters for subsequent control decisions, enabling the control system to distinguish between different types of anoxic conditions such as "insufficient gas supply" and "low mass transfer efficiency," laying a data foundation for refined and differentiated regulation.

[0073] In one specific implementation, generating operating frequency adjustment commands for the aeration equipment and start / stop control commands or speed adjustment commands for the auxiliary stirring equipment includes: when the gas-liquid coupling characteristic value is lower than a preset uniformity threshold, the system determines that the current aeration state has bubble aggregation or uneven distribution. First, the system continuously monitors the real-time gas-liquid coupling characteristic value output by intelligent fusion calculation and compares it with a set benchmark threshold; specifically, this benchmark threshold is determined statistically based on the gas-liquid coupling characteristic value under historical normal operating conditions, with a typical value range of 0.5 to 0.7, and a value of 0.6. Then, under the premise that the gas-liquid coupling characteristic value is lower than the threshold, it further determines whether the real-time collected dissolved oxygen concentration is lower than the lower limit of the dissolved oxygen concentration benchmark range. If the above two conditions are met, the control logic enters a collaborative response mode: at this time, the dual-channel control decision module generates commands to maintain the current blower frequency or appropriately reduce its operating frequency to avoid inputting excessive air into the pool, resulting in ineffective aeration and energy waste; simultaneously, it generates commands in parallel to start the underwater jet mixer or increase its speed to enhance the fluid shear force in the pool. Next, the shearing action of the water flow causes the aggregated large bubbles to break down into microbubbles, thereby increasing the gas-liquid contact area per unit volume of gas and improving oxygen mass transfer efficiency. In some other optional embodiments, the agitator speed can be adjusted using a segmented speed-up strategy, setting multiple speed-up levels based on the degree to which the gas-liquid coupling characteristic value deviates from the threshold; or, a closed-loop feedback mechanism can be used, using the rising gas-liquid coupling characteristic value in subsequent sampling cycles as an indicator of control effectiveness, dynamically adjusting the agitation intensity until it returns to normal levels.

[0074] Through the above-described scheme, this embodiment can accurately identify and differentiate interventions for inefficient aeration states. In cases of insufficient dissolved oxygen, it distinguishes between two causes: "insufficient air supply" and "mass transfer obstruction," and implements targeted control measures. Compared to existing technologies, this embodiment avoids the energy consumption increase problem caused by simply increasing air volume. By introducing a water flow shear-assisted mechanism, it effectively improves bubble dispersion and oxygen utilization efficiency, significantly reducing aeration energy consumption per unit of pollutant removal, while extending the service life of aeration equipment and enhancing the overall stability and energy efficiency of the wastewater treatment process.

[0075] In one specific implementation, the method further includes a condition attribution step based on correlation analysis: When the real-time dissolved oxygen concentration is lower than the dissolved oxygen concentration benchmark range, continuous time series data of ambient temperature and aeration equipment operating power are first acquired. Specifically, a linear regression fitting is performed on the ambient temperature series based on a sliding time window to calculate its slope within a preset time period to quantify the temperature rise trend; simultaneously, the same processing is performed on the operating power series to determine its corresponding slope. Then, it is determined whether the current operating condition meets the following combination of conditions: the slope of the ambient temperature change is positive, and the slope of the operating power change is less than or equal to zero. If this condition is met, the system is determined to be in a state of decreased oxygen mass transfer efficiency caused by the increase in ambient temperature, i.e., a thermal effect-dominated operating condition. Next, a frequency compensation command is generated for the aeration equipment to actively increase the blower operating frequency to offset the increase in mass transfer resistance caused by the decrease in saturated dissolved oxygen, thereby maintaining the oxygen supply level required for biochemical reactions. In some other optional implementations, the operating condition attribution logic can also combine atmospheric pressure change trends or seasonal meteorological parameters as auxiliary criteria; or a dynamic threshold mechanism can be used to adaptively adjust the slope judgment boundary based on the historical temperature change rate to improve attribution accuracy.

[0076] Through the above-described scheme, this embodiment can effectively identify the phenomenon of low dissolved oxygen measurement caused by rising ambient temperature, and distinguish the difference between this and the actual increase in oxygen demand, thus avoiding the control system misjudging it as a load surge and causing excessive gas supply. Compared with the prior art, this embodiment achieves intelligent identification and feedforward compensation of non-biological interference factors, significantly reducing unnecessary energy consumption fluctuations while ensuring the stability of wastewater treatment, and improving the environmental adaptability and operational efficiency of the control system.

[0077] In one specific implementation, the operating condition attribution step further includes: when the slope of the ambient temperature change is within a preset normal fluctuation range, the system continuously monitors the trend of water turbidity changes. Specifically, firstly, the slope of water turbidity change within a sliding time window of 60 seconds is calculated in real time. The least squares method is used to linearly fit the continuously sampled turbidity data to obtain the slope value at the current moment. Then, this slope value is compared with a preset abrupt change threshold, which is determined based on historical operating data statistics and ranges from 0.8 NTU / s to 1.5 NTU / s. If the turbidity change slope is detected to exceed this threshold, it is determined that the influent organic load has experienced a shock increase. Next, the system generates a feedforward control command to increase the operating frequency of the aeration equipment in advance before the dissolved oxygen concentration decreases significantly. The frequency increase is positively correlated with the turbidity change slope, and the proportional coefficient is determined through on-site calibration, with a typical value of 0.3 Hz / (NTU / s). In other alternative implementations, the mutation threshold can be dynamically adjusted according to the season or time period; or, an exponentially weighted moving average method can be used instead of linear regression to improve the response speed of slope calculation; or, the synchronous surge signal of influent flow can be combined as an auxiliary criterion to enhance the reliability of operating condition identification.

[0078] Through the above scheme, this embodiment can achieve proactive regulation when the influent organic pollutant load suddenly increases, effectively avoiding the risk of decreased treatment efficiency or effluent quality exceeding standards due to the delayed response of dissolved oxygen, and improving the anti-disturbance capability and control foresight of the sewage treatment system.

[0079] In one specific implementation, the method further includes a correction step for rainfall interference: real-time rainfall data is acquired through an external meteorological interface, and the operating condition is determined by combining the changing trends of influent flow rate and water turbidity. First, the central control unit periodically receives rainfall information from the meteorological service system, with a sampling interval of 1 minute; when the current rainfall exceeds a preset threshold (e.g., 5 mm / h), the rainwater dilution condition identification logic is activated. Specifically, the control unit simultaneously analyzes whether the influent flow rate shows an upward trend (e.g., an increase of more than 20% compared to the previous 10 minutes) and determines whether the water turbidity has decreased significantly (e.g., a decrease of more than 30%). If both conditions are met simultaneously, it is determined that the dilution effect is caused by a large amount of rainwater entering the pipe network. Next, the system dynamically corrects the dissolved oxygen concentration baseline range output by the neural network model: the lower limit of the originally calculated baseline range is multiplied by an attenuation coefficient α (in one implementation, α ranges from 0.7 to 0.9, with a typical value of 0.8), thereby reducing the target DO control lower limit. This adjustment prevents the blower from operating at a continuously high load due to maintaining an excessively high DO setting, thus reducing ineffective aeration, even when rainwater dilutes the pollutant load and the actual pollutant load decreases. In other optional implementations, the attenuation coefficient α can be set according to rainfall intensity levels, for example, α=0.9 for light rain (2–5 mm / h), α=0.8 for moderate rain (5–10 mm / h), and α=0.7 for heavy rain (>10 mm / h); alternatively, the correction method can also use a lookup table to determine the adjustment range of the baseline range based on historical statistical rainy season dilution response curves.

[0080] Through the above-described scheme, this embodiment can effectively identify influent dilution conditions caused by rainfall, proactively reduce the dissolved oxygen control target while ensuring the stability of biochemical treatment, and avoid excessive oxygen supply to the aeration system under low pollution load conditions, thereby significantly reducing unnecessary energy waste. Compared with existing technologies, this embodiment achieves intelligent adaptive response to external environmental disturbances, improves the energy efficiency optimization capability of the control system under complex dynamic conditions, and can further reduce overall aeration energy consumption by 8%–12%.

[0081] In one specific implementation, generating operating frequency adjustment commands for the aeration equipment includes optimization steps based on energy consumption ratio: First, the system collects energy consumption data and pollutant removal information per unit time in real time. The energy consumption data comes from the power meter output of the aeration equipment, and the pollutant removal amount is calculated using the difference in chemical oxygen demand (COD) between the influent and effluent and the flow rate. Specifically, the real-time energy efficiency ratio... Defined as:

[0082] This indicator reflects the energy consumed to remove one unit of pollutant under the current operating conditions. Then, a support vector machine regression (SVR) algorithm is used to construct the operating frequency adjustment. Changes in energy efficiency ratio The nonlinear mapping relationship between them. The training sample set consists of historical operating data, including experimental results of frequency perturbations under different operating conditions, to ensure that the model covers typical dynamic response characteristics. Then, under the premise of satisfying the dissolved oxygen concentration baseline range constraint, the gradient descent method is used to optimize the SVR model online, searching for the one that maximizes the real-time energy efficiency ratio. The minimized optimal frequency point is used as the target control value, and the corresponding frequency adjustment command is generated.

[0083] In other alternative implementations, the energy efficiency ratio can be calculated using an energy efficiency index based on biochemical oxygen demand (BOD) or ammonia nitrogen removal rate; the regression model can also use Gaussian process regression (GPR) or random forest regression (RFR) instead of support vector machine regression to adapt to the data distribution characteristics of wastewater treatment plants of different sizes; the optimization algorithm can also use heuristic methods such as particle swarm optimization (PSO) to achieve multivariate collaborative parameter tuning.

[0084] Through the above scheme, this embodiment can dynamically identify the optimal operating point, minimizing unit treatment energy consumption and improving overall operational economy while ensuring stable dissolved oxygen levels. Compared with existing technologies, this embodiment avoids the over-aeration problem caused by unidirectional frequency amplification based solely on DO deviation, achieving a refined evolution of the control objective from "stability priority" to "energy saving priority," and the measured comprehensive energy consumption can be further reduced by 8%-12%.

[0085] In one specific implementation, the method further includes a device health status monitoring step: the central control unit is further configured to collect mechanical vibration signals during the operation of the aeration equipment to achieve real-time monitoring of the equipment's health status. Specifically, the system acquires vibration data through a triaxial accelerometer installed on the housing of the air-suspended blower, with the sampling frequency set to 1000 Hz to ensure coverage of the fundamental frequency and its main harmonic components. First, the acquired raw vibration signals are preprocessed, including de-meaning and bandpass filtering (0.5–500 Hz), to eliminate static gravity components and environmental interference noise. Then, a Fast Fourier Transform (FFT) is used to convert the time-domain vibration signal to the frequency domain to obtain its spectral distribution. Next, the fundamental frequency of the blower rotor (corresponding to the current operating frequency) and characteristic harmonic components such as the second and third harmonics are identified in the spectrum, and the amplitude energy at each frequency point is calculated.

[0086] When the amplitude of any characteristic frequency exceeds a threshold determined based on historical health data (e.g., exceeding the normal operating average plus two standard deviations), the system determines that the equipment has a potential mechanical failure risk, such as dynamic imbalance or bearing wear. In this case, during the generation of subsequent frequency adjustment commands, the system automatically limits the frequency rise rate, dynamically adjusting it from the default maximum rise rate (e.g., 5 Hz increase per second) to a safe rate (e.g., 1 Hz increase per second or remaining unchanged), thereby avoiding resonance or exacerbating mechanical damage due to rapid acceleration. This rate-limiting strategy remains in effect until the vibration amplitude falls below the threshold and remains stable for a period of time, after which the normal control logic resumes.

[0087] In other alternative implementations, vibration signal analysis can use wavelet packet transform instead of FFT to improve time-frequency resolution under non-stationary conditions; or introduce kurtosis index as an auxiliary criterion to enhance sensitivity to impact faults; or remote data access can be achieved through wireless vibration sensing modules, which is suitable for distributed deployment scenarios.

[0088] Through the above solution, this embodiment can effectively identify early mechanical abnormalities of aeration equipment and actively suppress aggressive frequency adjustment behavior at the control level, preventing further damage to the equipment in a deteriorated state, improving the operational safety of the system and the service life of the equipment, while avoiding the risk of uncontrolled sewage treatment process due to sudden shutdown.

[0089] In one specific implementation, the method further includes a time-segmented control step based on K-means clustering: First, daily periodic features are extracted from the historical operating data of the wastewater treatment system, specifically including the hourly average influent flow rate, average turbidity, average deviation between the setpoint and actual dissolved oxygen values, and average power consumption of aeration equipment, forming a multidimensional time series feature vector. Then, using the time points of the entire 24-hour period as input samples, the K-means clustering algorithm is used to perform unsupervised classification on the feature vector, with a preset cluster size of 3, resulting in three significantly different time period categories: low load period (e.g., 02:00–05:00), medium load period (e.g., 06:00–10:00, 18:00–21:00), and high load period (e.g., 11:00–17:00). Next, a confidence threshold for the neural network model output is configured for each time period category: in the low-load time period category, the confidence threshold is set to a lower level (e.g., 0.1) to suppress frequent control actions caused by minor disturbances; in the high-load time period category, the confidence threshold is raised to a higher level (e.g., 0.3) so that a control response is triggered immediately when the predicted output changes beyond this threshold, thereby enhancing the system's dynamic adjustment capability. Specifically, the confidence threshold is adaptively adjusted based on the error statistics and process response delay characteristics accumulated during online learning.

[0090] In some other alternative implementations, the clustering algorithm can be replaced with a Gaussian mixture model (GMM) or a hierarchical clustering method to adapt to the continuous changes in the operating modes of different plant areas; the number of time period categories can be expanded to 2 to 5 categories according to the actual working conditions; the confidence threshold adjustment strategy can also be based on a fuzzy logic controller to achieve continuous mapping rather than discrete step switching.

[0091] Through the above scheme, this embodiment can achieve differentiated control sensitivity management based on the daily cycle load fluctuation characteristics of the sewage treatment process. During the low load period, it reduces the frequency of unnecessary actuator actions, extends equipment life and reduces energy consumption. During the high load period, it improves the response speed and stability of the control system, thereby optimizing the overall operating efficiency while ensuring the quality of the effluent.

[0092] In one specific implementation, the method further includes a linkage control step for the mud-water mixing state: The sludge loading parameters are calculated based on real-time collected water turbidity and influent flow rate data. First, the system obtains the influent flow rate per unit time through a multi-dimensional sensing interface module. (Unit: m³ / h) and the average turbidity for the corresponding time period (Unit: NTU), and combined with preset empirical coefficients. (Unit: mg / L / NTU, reflecting the conversion relationship between turbidity and suspended solids concentration), estimate the approximate concentration of organic pollutants in the influent. Specifically, sludge loading parameters The dynamic calculation is performed according to the following formula:

[0093] in, The concentration of suspended solids in the mixed liquor in the biological treatment tank (MLSS, unit: mg / L) is provided by historical operating data or online MLSS sensor; This refers to the effective volume of the aeration tank (unit: m³). Then, the system determines the current... Does the value exceed the preset high load threshold? (For example, 0.35 kgBOD / kgMLSS·d). If this is exceeded, the influent concentration is considered too high, posing a risk of sludge load imbalance. Next, the dual-channel control decision module generates adjustment commands, increasing the opening of the sludge return pump via the drive interface to improve the sludge return ratio. More activated sludge is returned from the secondary sedimentation tank to the front end of the biological treatment tank to increase the total number of microorganisms in the system and maintain... Within a reasonable range. This linkage control is synchronized with the frequency adjustment of the aeration equipment to ensure that the oxygen supply capacity matches the biodegradation requirements.

[0094] In some other alternative implementations, the calculation of sludge load parameters can further incorporate measured data from an online COD monitor to replace the turbidity conversion value, thereby improving the accuracy of load estimation; alternatively, the adjustment of the return pump flow rate can employ a proportional-integral (PI) control algorithm, based on... The deviation is dynamically output as a control signal, rather than being adjusted by switching based on a fixed threshold.

[0095] Through the above-described scheme, this embodiment can achieve dynamic balance control of the sludge-water mixing state in the wastewater treatment system. Compared with the prior art, this embodiment, when the influent load suddenly increases, not only increases the aeration rate to meet the oxygen demand, but also simultaneously adjusts the sludge return to ensure a sufficient microbial population, avoiding a decrease in treatment efficiency due to "aerobic but sterile" or "sterile but anaerobic" conditions, thereby improving the shock resistance and operational stability of the entire biological system.

[0096] In one specific implementation, the method further includes an online model update step: The multi-source time-series data generated during daily operation, along with the corresponding control execution results, are continuously written into the historical database. Specifically, the system is first configured with a timestamped data archiving module, which stores a complete data vector, including dissolved oxygen concentration, turbidity, flow rate, temperature, power, pressure pulsation signals, and the blower frequency and mixer speed at the corresponding time, in a structured manner on a daily basis.

[0097] When the preset update cycle (e.g., every 7 days) arrives, the model update process is triggered: new running data fragments are extracted from the database, and the data is preprocessed in the same way as in the initial training phase, including time alignment, denoising, and feature construction.

[0098] An incremental learning strategy is employed to fine-tune the parameters of the deployed Long Short-Term Memory (LSTM) network model. This involves using the original model weights as the initialization point and performing several rounds of backpropagation training with only new data, thus avoiding catastrophic forgetting. After training, the new model automatically replaces the old model running online once its prediction accuracy is evaluated on a validation set.

[0099] In other alternative implementations, incremental training can employ a sliding time window mechanism, retaining only the most recent 90 days of valid running data for retraining to enhance the model's adaptability to recent operating conditions; or an online learning framework can be introduced, such as using mini-batch gradient descent to update parameters batch by batch, further improving the model's real-time response to environmental changes.

[0100] Through the above scheme, this embodiment can continuously absorb dynamic data in actual operation, enabling the neural network model to gradually adapt to the long-term characteristic drift of the sewage treatment system caused by factors such as seasonal changes, microbial community succession, or equipment aging, thereby maintaining the stability of prediction accuracy and control performance, extending the effective service life of the model, and reducing the need for manual intervention.

[0101] This invention provides a multi-source data fusion modeling system for an aeration process. For example... Figure 2 As shown, the system includes: The multi-dimensional sensing interface module M100 is configured to establish a bidirectional connection with a group of field sensors distributed in the wastewater treatment tank via an industrial communication bus. The field sensor group includes at least a dissolved oxygen sensor, a turbidity meter, an influent flow meter, a temperature sensor, a power meter, and a high-frequency pressure sensor installed at the bottom of the aeration tank and located 30 to 50 centimeters above the aeration head. The multi-dimensional sensing interface module M100 synchronously acquires the time-series signals output by the above sensors at a sampling period of 1 second and packages them into a structured data stream for use by subsequent processing units.

[0102] The timing synchronization processor M200, electrically connected to the multi-dimensional sensing interface module M100, is configured to cache multi-source time series data within a continuous time window. The processor executes a cross-correlation analysis algorithm, selects the running power sequence as the driving benchmark, calculates the maximum correlation offset between it and the dissolved oxygen concentration sequence, and thus determines the lag time in the physical response. Based on this lag time parameter, the processor performs resampling and timestamp alignment operations on the original dataset to generate a calibration dataset that remains synchronized in the dynamic response, and outputs it to the downstream computing unit.

[0103] The intelligent fusion computing core M300 includes non-volatile memory and a high-performance central processing unit. The memory pre-stores a neural network model trained with historical operating data. The processor is configured to receive a calibration dataset from the timing synchronization processor M200, extract water turbidity, influent flow rate, and ambient temperature as input feature vectors, and feed them into the neural network model for inference calculations, outputting the corresponding dissolved oxygen concentration benchmark range under the current operating conditions. At the same time, the processor performs frequency domain transformation processing on the received high-frequency pressure signal segments, and extracts gas-liquid coupling feature values ​​to characterize the bubble morphology by integrating the energy distribution of different frequency segments. These feature values ​​reflect the size uniformity and breakup state of bubbles in the current gas-liquid two-phase flow.

[0104] The dual-channel control decision module M400, connected to the intelligent fusion computing communication, is configured to receive dissolved oxygen reference range, real-time dissolved oxygen measurement value, and gas-liquid coupling characteristic value. When the gas-liquid coupling characteristic value deviates from the normal range, it generates start / stop control commands or speed adjustment commands for the auxiliary stirring equipment. When the real-time dissolved oxygen concentration is continuously lower or higher than the set reference range, it generates frequency adjustment commands for the frequency converter controller of the aeration equipment. This module dynamically adjusts the control priority and response intensity of the two types of actuators according to different combinations of monitoring states.

[0105] The M500 actuator interface is connected to the frequency converter of the air-suspended blower and the motor driver of the underwater jet mixer via digital communication links and analog output channels, respectively. It is used to convert the digital control commands generated by the control decision module into executable electrical signals and send them to the corresponding actuators in real time, so as to achieve coordinated control of aeration intensity and water flow shear force.

[0106] In one embodiment, the high-frequency pressure sensor employs a piezoresistive sensing element, featuring a sampling bandwidth of ≥500Hz and an IP68 protection rating, ensuring long-term stable operation in high humidity and highly corrosive environments. In other alternative embodiments, capacitive or fiber optic Fabry-Perot interferometric pressure sensors can be used instead to adapt to different signal-to-noise ratio requirements or environments with strong electromagnetic interference.

[0107] In one embodiment, the neural network model employs a Long Short-Term Memory (LSTM) network architecture, which can capture the dynamic evolution trend of operating parameters; in other alternative embodiments, a hybrid model combining a gated recurrent unit (GRU) network or a one-dimensional convolutional neural network (1D-CNN) with an attention mechanism can also be used to achieve similar time series prediction functions.

[0108] In one embodiment, the extraction of gas-liquid coupling features is based on spectral decomposition using Fast Fourier Transform (FFT); in other alternative embodiments, wavelet packet transform or multi-resolution analysis methods can also be used for time-frequency joint analysis to improve the sensitivity of identifying transient bubble bursting events.

[0109] The above modules form a closed-loop collaborative relationship: the multi-dimensional sensing interface module M100 provides raw observation data, the timing synchronization processor M200 eliminates the data asynchrony problem caused by sensor response differences, and makes the data input to the model conform to the real physical causal order; the intelligent fusion calculation uses the calibrated data to simultaneously complete the two tasks of "oxygen demand prediction" and "mass transfer status perception", and outputs control targets and diagnostic basis respectively; the dual-channel control decision module M400 combines these two types of information to determine whether the root cause of hypoxia is "insufficient air supply" or "mass transfer failure", and then selects different strategies such as adjusting the blower or starting the agitator; finally, the command is converted into physical action through the execution drive interface M500, which acts on the process system, changes the aeration effect and feeds back to the sensor, forming a complete perception-analysis-decision-execution loop.

[0110] This multi-level, cross-modal information fusion mechanism effectively solves the misadjustment problem caused by the lag of DO sensors in traditional control systems, and avoids the energy waste caused by blindly increasing ventilation based solely on a single dissolved oxygen index. At the same time, under the condition that the behavior of bubbles inside turbid wastewater cannot be observed visually, the gas-liquid contact efficiency is indirectly evaluated through pressure pulsation spectrum, realizing the transparency of the "black box" mass transfer process. Combined with the differentiated response strategy of dual actuators, the system can take targeted intervention measures under different fault modes, significantly improving control accuracy and energy efficiency.

[0111] Through the above scheme, this embodiment can realize online identification and fine control of gas-liquid mass transfer state during sewage treatment, reduce energy consumption caused by ineffective aeration while ensuring stable effluent quality, and improve the operational robustness and adaptability of the entire biochemical treatment system.

[0112] In one specific implementation, the intelligent fusion computing further integrates a physical constraint logic unit. This physical constraint logic unit is configured to store a dissolved oxygen saturation calculation model based on Henry's Law and dynamically calculate the theoretical saturated dissolved oxygen threshold under the current operating conditions by receiving real-time ambient temperature data. The physical constraint logic unit is connected to the output of the neural network model. Upon receiving the dissolved oxygen baseline interval calculated by the neural network, it immediately performs a compliance check on the upper limit of the baseline interval. When it detects that the upper limit exceeds the theoretical saturated dissolved oxygen threshold calculated based on the current water temperature, the physical constraint logic unit triggers a signal truncation mechanism, forcibly clamping the upper limit of the original output baseline interval to the theoretical saturated dissolved oxygen threshold, thereby generating a corrected dissolved oxygen control target interval that conforms to physical laws and transmitting it to the dual-channel control decision module M400 for subsequent instruction generation.

[0113] Henry's Law calculation model is implemented using empirical formulas:

[0114] in The water temperature (unit: °C) is collected in real time. The local real-time atmospheric pressure is given in kPa. This formula is embedded in the read-only memory area of ​​the physical constraint logic unit using a lookup table or a pre-defined function mapping, ensuring rapid threshold determination without external dependencies.

[0115] In some other alternative implementations, the saturated dissolved oxygen calculation in the physical constraint logic unit can be replaced by a multidimensional interpolation lookup table, which is pre-calibrated according to different temperature and salinity combinations and stored in non-volatile memory; or, the signal truncation mechanism can be configured as a graded response mode, issuing only an alarm signal when the predicted value exceeds the saturation limit by less than 10%, and performing a forced clamping operation only when it exceeds 10%; or, the physical constraint logic unit can also be extended to support the rationality check of the input feature vector, for example, automatically enabling the default safety value to participate in the calculation when the temperature input exceeds the normal range of 0–40℃.

[0116] Through the above solution, this embodiment can effectively prevent the problem of dissolved oxygen setpoints violating thermodynamic limits due to prediction biases in neural network models, and avoid the control system outputting invalid or even erroneous frequency adjustment commands. Compared with the prior art, this embodiment introduces rigid physical boundary constraints through hardware-level logic units without sacrificing model learning capabilities, significantly improving the safety and engineering reliability of the AI-driven aeration control system, and is particularly suitable for actual wastewater treatment scenarios with frequent water quality disturbances and large fluctuations in operating conditions.

[0117] In one specific implementation, the dual-channel control decision module M400 embeds collaborative optimization logic. This logic is configured to receive gas-liquid coupling characteristic values ​​from the intelligent fusion calculation output and compare them with a preset uniformity threshold in real time. When the gas-liquid coupling characteristic value is detected to be continuously lower than the threshold for more than a set time window (e.g., five consecutive sampling cycles), it is determined that the bubble breakage in the current aeration tank is insufficient, posing a risk of large bubble aggregation or local blockage of the aeration device. Under this condition, the collaborative optimization logic immediately activates the dual-channel linkage mechanism: on the one hand, by reducing or locking the gain coefficient of the frequency adjustment command sent to the frequency converter controller of the aeration equipment, it suppresses further increases in the air supply of the blower, preventing ineffective over-aeration; on the other hand, it generates a positive speed increase command in parallel, which is transmitted to the driver of the underwater jetting and mixing equipment via the execution drive interface M500 to enhance the fluid shear force in the tank, promote the breakup of large bubbles into microbubbles, thereby improving the gas-liquid contact area and oxygen mass transfer efficiency. This coordinated action continues until the gas-liquid coupling characteristic value rises back to the preset threshold range and remains stable for no less than 30 seconds. At this point, it is determined that the bubble distribution has returned to normal, and the system resumes the normal dual-channel independent control mode.

[0118] In some other alternative implementations, the preset uniformity threshold can be dynamically adjusted according to seasonal water quality changes or different wastewater treatment process sections (such as aerobic zones and anoxic zones), and determined by a lookup table method or a sliding percentile based on historical data; the gain lock strategy can be replaced by segmented amplitude limiting control, that is, setting multi-level frequency adjustment upper limits according to the deviation of the gas-liquid coupling characteristic value; the parallel speed increase command can also be output in a gradient increment manner, with an initial step increase of 10% of the rated speed, and then an increment of 5% every 15 seconds until the optimal shear strength or trigger current protection limit is reached.

[0119] Through the above-described scheme, this embodiment can achieve accurate identification and rapid response to adverse bubble morphology conditions, and complete closed-loop control of gas-liquid mass transfer without relying on visual perception. By limiting inefficient gas supply and actively enhancing mechanical disturbances, it avoids the energy waste and dissolved oxygen overshoot problems caused by blindly increasing airflow in traditional control, and significantly improves the dynamic adaptability and operational efficiency of the aeration system.

[0120] In one specific implementation, the timing synchronization processor M200 integrates an adaptive Kalman filter, which is configured to perform recursive filtering on the resampled and aligned calibration dataset. Specifically, a first-order linear state-space model is constructed with the actual dissolved oxygen concentration as the system state variable, and its state transition equation is expressed as follows: ,in Indicates the first The actual dissolved oxygen concentration at any given time. Assuming the process noise follows a zero-mean Gaussian distribution, the observation equation is defined as follows: ,in These are the actual measured values ​​from the sensor. To mitigate measurement noise, the filter dynamically adjusts the Kalman gain by estimating the covariance matrix of process noise and measurement noise in real time. This suppresses high-frequency random noise while preserving the signal's phase characteristics, avoiding the introduction of additional hysteresis. The filtered dissolved oxygen sequence is then transmitted as a high signal-to-noise ratio (SNR) input vector to the intelligent fusion computation for subsequent modeling and prediction. In other alternative implementations, the adaptive Kalman filter can be replaced by a maximum likelihood estimation noise parameter update mechanism based on a sliding window, or a square root capacitive Kalman filter (SR-CKF) structure can be used to improve numerical stability, particularly suitable for dissolved oxygen signal processing in low SNR or non-Gaussian noise environments.

[0121] Through the above solution, this embodiment can effectively reduce the impact of random noise caused by electrode aging, water flow disturbance or electrical interference during dissolved oxygen sensor measurement, significantly improve the quality of input data without sacrificing response speed, and thus enhance the stability of neural network model prediction and the accuracy of control decisions.

[0122] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A multi-source data fusion modeling method for an aeration process, characterized in that, The method is applied to the aeration stage of a wastewater treatment system and includes the following steps: Collect multi-source time series data during the wastewater treatment process. The multi-source time series data includes dissolved oxygen concentration, water turbidity, influent flow rate, ambient temperature, operating power of aeration equipment, and liquid pressure pulsation signal at the bottom of the tank. The change sequence of the operating power is selected as a reference benchmark. The time lag parameter of the dissolved oxygen concentration relative to the operating power is calculated. The multi-source time series data is then time-series aligned and corrected according to the time lag parameter to obtain the calibrated multi-source time series data. A neural network model for predicting oxygen demand is constructed. The water turbidity, influent flow rate and ambient temperature in the calibrated multi-source time series data are used as input features to the neural network model, and the model outputs the baseline range of dissolved oxygen concentration under the current operating conditions. The spectral features of the liquid pressure pulsation signal at the bottom of the pool are extracted to obtain gas-liquid coupling feature values ​​that characterize the morphological distribution of bubbles in wastewater. Based on the dissolved oxygen concentration benchmark range, the dissolved oxygen concentration collected in real time and the gas-liquid coupling characteristic value are used to generate operating frequency adjustment commands for the aeration equipment and start / stop control commands or speed adjustment commands for the auxiliary stirring equipment, so as to maintain dissolved oxygen stability in the wastewater treatment process.

2. The multi-source data fusion modeling method for the aeration process according to claim 1, characterized in that, The time lag parameter for calculating the dissolved oxygen concentration relative to the operating power includes: Extract the operating power sequence and dissolved oxygen concentration sequence within a historical time window of a preset length; Calculate the cross-correlation function between the operating power sequence and the dissolved oxygen concentration sequence; Search for the peak value of the cross-correlation function, and determine the time offset corresponding to the peak value as the time lag parameter; The time lag parameter is shifted forward by the timestamp of the dissolved oxygen concentration sequence to establish a calibration dataset for physical response synchronization.

3. The multi-source data fusion modeling method for the aeration process according to claim 1, characterized in that, After obtaining the calibrated multi-source time series data, the method further includes: A state-space equation is constructed, with the actual dissolved oxygen concentration set as the system state variable and the sensor measurement noise set as the observation noise. The dissolved oxygen concentration data in the calibrated multi-source time series data is iteratively estimated using the extended Kalman filter algorithm, and high-frequency random noise is filtered out while preserving the signal phase characteristics.

4. The multi-source data fusion modeling method for the aeration process according to claim 1, characterized in that, The neural network model adopts a long short-term memory network architecture, and the construction process of the neural network model includes: Extract the water turbidity change rate and influent flow rate change rate from historical operational data as dynamic feature vectors; The dissolved oxygen concentration at historical moments is used as an autoregressive feature vector; The long short-term memory network is trained using the historical operating data, and the mean square error between the predicted output and the actual dissolved oxygen concentration is minimized through the backpropagation algorithm to establish the network weight parameters.

5. The multi-source data fusion modeling method for the aeration process according to claim 4, characterized in that, The training process of the neural network model also includes a physical constraint correction step: The theoretical saturated dissolved oxygen concentration was calculated using Henry's Law based on the real-time collected ambient temperature. A penalty term is added to the loss function of the neural network model. When the predicted value output by the neural network model exceeds the theoretical saturated dissolved oxygen concentration, the loss function value is increased to limit the output of the neural network model to conform to physical laws.

6. The multi-source data fusion modeling method for the aeration process according to claim 1, characterized in that, The step of extracting spectral features from the pressure pulsation signal at the bottom of the pool includes: The pressure pulsation signal at the bottom of the pool is subjected to a fast Fourier transform to convert the time-domain signal into a frequency-domain power spectrum. The ratio of the energy value in the preset high-frequency range to the total energy value of the entire frequency band is calculated by integral calculation. The ratio is determined as the gas-liquid coupling characteristic value, which is used to characterize the uniformity of bubble size generated by aeration.

7. The multi-source data fusion modeling method for the aeration process according to claim 6, characterized in that, The generation of the operating frequency adjustment command for the aeration equipment and the start / stop control command or speed adjustment command for the auxiliary stirring equipment includes: When the gas-liquid coupling characteristic value is lower than the preset uniformity threshold, the aeration state is determined to be bubble aggregation or uneven distribution. In the aeration state, if the real-time collected dissolved oxygen concentration is lower than the lower limit of the dissolved oxygen concentration reference range, an instruction is generated to maintain or reduce the operating frequency of the aeration equipment, and an instruction to start the auxiliary stirring equipment or increase the speed of the auxiliary stirring equipment is generated in parallel to break up the polymerized bubbles using the shear force of the water flow.

8. The multi-source data fusion modeling method for the aeration process according to claim 1, characterized in that, The method also includes a condition attribution step based on correlation analysis: When the real-time collected dissolved oxygen concentration is lower than the dissolved oxygen concentration reference range, calculate the slope of the change in ambient temperature and the slope of the change in operating power; If the slope of the change in ambient temperature is positive and the slope of the change in operating power is not positive, it is determined that the oxygen mass transfer coefficient is reduced due to the environmental thermal effect, and a temperature compensation command to increase the operating frequency of the aeration equipment is generated.

9. The multi-source data fusion modeling method for the aeration process according to claim 8, characterized in that, The condition attribution step also includes: If the slope of the change in ambient temperature is within a preset range, but the slope of the change in water turbidity exceeds the abrupt change threshold, it is determined to be an influent organic load shock. Generate feedforward control commands to increase the operating frequency of the aeration equipment before the dissolved oxygen concentration actually decreases.

10. A multi-source data fusion modeling system for an aeration process, characterized in that, The system includes: The multidimensional sensing interface module is configured to communicate with a field sensor group distributed in the sewage treatment tank. The field sensor group includes a dissolved oxygen sensor, a turbidity meter, a flow meter, a temperature sensor, a power meter, and a high-frequency pressure sensor installed at the bottom of the tank. The multidimensional sensing interface module is used to receive multi-source time series data in real time. A timing synchronization processor is configured to cache the multi-source time series data, execute a cross-correlation analysis algorithm to determine the response lag time of dissolved oxygen concentration relative to operating power, resample and align the multi-source time series data based on the response lag time, and output a physically synchronized calibration dataset. The intelligent fusion computing core includes a memory and a processor. The memory stores a pre-trained neural network model. The processor is configured to: input the water turbidity, influent flow rate and ambient temperature from the calibration dataset into the neural network model, calculate the dissolved oxygen reference range under the current operating conditions, and perform frequency domain transformation on the signal from the high-frequency pressure sensor to extract gas-liquid coupling feature values ​​to characterize the physical morphology of bubbles. The dual-channel control decision module is configured to receive the output results of the intelligent fusion computing core. When the gas-liquid coupling characteristic value is abnormal, it generates start / stop control commands or speed adjustment commands for the auxiliary stirring equipment; when the real-time dissolved oxygen concentration deviates from the dissolved oxygen reference range, it generates frequency adjustment commands for the aeration equipment. The execution drive interface is connected to the frequency converter controller of the aeration device and the driver of the auxiliary stirring device, respectively, and is used to issue the frequency adjustment command and the speed adjustment command.

Images