Wastewater treatment aerobic space-aeration double-loop control method based on nitrification process

Abstract

This invention relates to the field of wastewater treatment technology, specifically to a dual-loop control method for aerobic space-aeration based on the nitrification process in wastewater treatment, constructing a dual loop of "on-demand aeration" and "on-demand aerobic" aeration. In the aeration loop, a SAD (Self-Aeration Optimization) method is proposed. n The indicator integrates deep learning feedforward prediction and PID feedback correction with dead zone to precisely control aeration rate directly with ammonia nitrogen removal as the target; in the aerobic space loop, a SAD is established. n A U-shaped curve correlation model with aeration intensity is used to solve for the economic aeration intensity through a neural network, and the optimal aerobic zone volume is dynamically calculated in conjunction with the target aeration rate. The two loops are decoupled and coordinated based on the nitrification process, ensuring the aeration system always operates at its optimal energy efficiency point. This invention achieves in-situ, zero-delay, and precise control of the biological nitrification process. While ensuring stable compliance with effluent ammonia nitrogen standards, it significantly reduces aeration energy consumption and the unit consumption of added carbon sources, improves total nitrogen removal rate, and achieves energy saving, consumption reduction, and quality improvement in wastewater treatment.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, specifically to a dual-loop control method for aerobic space-aeration based on the nitrification process in wastewater treatment. Background Technology

[0002] Currently, municipal wastewater treatment plants generally adopt the activated sludge biological treatment process, in which the aeration system is the core of the biological treatment. Due to long-standing technical limitations of online ammonia nitrogen and other water quality instruments, the wastewater treatment industry has been forced to use dissolved oxygen as a substitute parameter for aeration control. However, in reality, dissolved oxygen is a relatively broad environmental parameter, with its control range often between 1-4 mg / L. It does not have a strictly linear relationship with the ultimate goal of ammonia nitrogen removal, which leads to relatively crude control in actual production, easily causing over-aeration and a large waste of energy.

[0003] Meanwhile, the actual influent water quality and quantity of most wastewater treatment plants are far below design values, yet they still operate using the designed biological tank functional zones and operating parameters. This results in the aerobic tank volume remaining redundant for extended periods. Under low-load conditions, this redundant tank volume cannot be effectively identified and utilized, further exacerbating the over-aeration problem. Over-aeration not only damages the sludge floc structure and reduces sludge activity but also leads to a surge in electricity costs. Furthermore, it wastes raw water carbon sources, resulting in low efficiency in denitrification and biological phosphorus removal. Consequently, additional carbon sources are often required, increasing operating costs and carbon emissions.

[0004] Existing control methods fail to achieve a precise match between aeration volume and aerobic space, based on the fundamental principle of "on-demand supply." Therefore, how to break through the control model focused solely on dissolved oxygen and establish a control method that can dynamically and precisely adjust aeration volume and aerobic zone volume according to the real-time nitrification process, thereby maximizing energy conservation and improving nitrogen and phosphorus removal efficiency while ensuring stable effluent compliance, has become a pressing technical challenge in this field. Summary of the Invention

[0005] The purpose of this invention is to provide a dual-loop control method for aerobic space-aeration based on the nitrification process in wastewater treatment, in order to solve the problems of high energy consumption, low pollutant treatment efficiency and low carbon source utilization caused by excessive aeration due to the reliance on dissolved oxygen for indirect control and the solidification of functional zones in biological ponds in existing technologies.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a wastewater treatment aerobic space-aeration dual-loop control method based on nitrification process, including an on-demand aeration loop and an on-demand aerobic space loop;

[0007] The on-demand aeration loop acquires the influent flow rate, influent ammonia nitrogen concentration, and effluent ammonia nitrogen concentration in the aeration zone in real time, and calculates the nitrification ratio air demand (SAD).n SAD n The actual amount of air consumed to remove a unit mass of ammonia nitrogen; based on SAD n Based on the target ammonia nitrogen removal load, predict the required aeration volume, and combine feedback correction, accurately calculate the target aeration volume and control the aeration equipment to supply air as needed; Establish SAD on demand aerobic space loop n The correlation model with aeration intensity g, where aeration intensity g is the ratio of the total aeration volume to the effective volume of the aeration zone, is used to solve for SAD. n Minimum economic aeration intensity g opt Based on g opt Calculate the optimal aerobic zone volume V based on the target aeration rate. req V req Compared with the current actual aerobic zone volume V current Compare and dynamically adjust the volume of the aerobic zone based on the comparison results; The on-demand aeration loop and the on-demand aerobic space loop are decoupled based on the nitrification process and controlled collaboratively.

[0008] Furthermore, based on the precise integral model for tracking microscopic water masses, a sliding window accumulation is performed from the microscopic to the macroscopic level to obtain the SAD. n The calculation formula is as follows: (Nm) 3 / kgN); Where G is the total aeration volume, Q is the influent flow rate of the aeration zone, and N is the total aeration volume. in The ammonia nitrogen concentration in the influent of the aeration zone, N out t represents the ammonia nitrogen concentration in the effluent from the aeration zone, w represents the sliding window time, and HT represents the time it takes for the mixed liquor to pass through the aeration zone.

[0009] Furthermore, a feedforward control algorithm is used to predict the required aeration rate. The calculation formula for the feedforward control algorithm is as follows: (Nm) 3 / h) Among them, G a,req To predict the required aeration rate, Q in N is the influent flow rate to the aeration zone. in The ammonia nitrogen concentration in the influent of the aeration zone, N tar The target effluent ammonia nitrogen concentration for the aeration zone.

[0010] Furthermore, the SAD mentioned above n The optimal nitrification ratio gas demand (SAD) predicted by the deep learning model n,optimal (t); Deep learning models are used to learn the mapping relationship f: X(t) → SAD from real-time measurable operating condition features X(t). n,optimal(X(t)); Operating condition characteristics X(t) include historical SAD. n One or more of the following: influent flow rate, biological tank water temperature, ammonia nitrogen concentration in the influent and effluent of the aerobic zone, volume of the aerobic zone, MLVSS, DO, and time; The formula for predicting the required aeration volume is: (Nm) 3 / h).

[0011] Furthermore, the feedback correction employs a PID control algorithm with dead zone; The aforementioned PID control algorithm with dead zone has a deviation e dead (t) is defined as:

[0012] in, This indicates a deviation in the control of ammonia nitrogen in the effluent. This is the preset dead zone width; Target aeration rate G a,set (k) Predict the required aeration volume G from the feedforward. a,d (k) is formed by superimposing the PID feedback correction amount u(k): G a,set (k)=G a,d (k)+u(k).

[0013] Furthermore, in the aforementioned on-demand aerobic spatial loop, a neural network model is constructed to learn the gas demand for nitrification ratio (SAD) from aeration intensity g and operating condition characteristics X. n continuous mapping f θ (g,X); By introducing a physical prior regularization term into the loss function, it is ensured that for any operating condition X, the mapping f θ (g,X) is a strictly convex function within the feasible region of aeration intensity, i.e., it satisfies the U-shaped curve characteristic; by solving for the global minimum point of the strictly convex function, the economic aeration intensity g under the current operating conditions can be obtained. opt (X).

[0014] Furthermore, the loss function includes: Mean squared error term used to ensure the accuracy of model predictions: ; Physical regularization terms used to enforce U-shaped convexity constraints: ; in, To predetermine the lower bound of curvature, The regularization strength; Boundary gradient constraints used to guide the minimum value to fall within the feasible region: .

[0015] Furthermore, the optimal aerobic zone volume The calculation formula is:

[0016] in, The volume requirement for the aerobic zone is based on the aeration intensity; Target aeration rate; Minimum volume requirement based on nitration kinetics school core; This is the absolute minimum volume of the aerobic zone in the process.

[0017] Furthermore, the dynamic adjustment of the aerobic zone size adopts a step-by-step adjustment strategy, specifically including: Configure and adjust the dead zone ratio and time lock (TIock); When the deviation ratio Exceeding the dead zone ratio An adjustment is triggered when the time since the last adjustment exceeds the time lock Tlock. like > Turn off the corresponding number of aeration units at the end of the aerobic zone to switch it to the anoxic zone; if Start the corresponding number of aeration units at the end of the aerobic zone to switch it to the aerobic zone.

[0018] The control system based on the above method includes a data acquisition module, an on-demand aeration control module, an on-demand aerobic space control module, and a core processing module; The data acquisition module is used to collect process parameters in real time, including influent flow rate in the aeration zone, ammonia nitrogen concentration in the influent of the aeration zone, ammonia nitrogen concentration in the effluent of the aeration zone, water temperature in the biological tank, MLVSS, and DO. The on-demand aeration control module includes SAD n The calculation unit, feedforward prediction unit, and feedback correction unit are used to execute the on-demand aeration loop, generate the target aeration volume, and send it to the aeration actuator. The on-demand aerobic space control module includes an association model construction unit, an economic aeration intensity solution unit, an optimal aerobic zone volume calculation unit, and a space adjustment decision unit, which are used to execute the on-demand aerobic space loop, generate space adjustment instructions, and send them to the aeration valve or control terminal. The core processing module is used to run deep learning algorithms, store historical data, and coordinate the collaborative work of the on-demand aeration control module and the on-demand aerobic space control module.

[0019] The beneficial effects of this invention are: 1. This invention utilizes the nitrification ratio gas demand (SAD) n By combining indicators and deep learning prediction models, an on-demand aeration method is established with ammonia nitrogen removal as the direct target, achieving dynamic and precise supply and demand balance of aeration volume, fundamentally eliminating over-aeration, significantly reducing aeration power consumption, and significantly saving energy and reducing consumption.

[0020] 2. This invention constructs SAD n The U-shaped curve correlation model with aeration intensity dynamically solves the economic aeration intensity and intelligently adjusts the volume of the aerobic zone accordingly, so that the aeration system always operates at the optimal energy efficiency point, simultaneously improving oxygen transfer efficiency and tank volume utilization, realizing the ultimate efficient utilization of aerobic space, and improving system energy efficiency.

[0021] 3. This invention converts the redundant aerobic zone volume into an anoxic zone in a timely manner, prolongs the denitrification reaction time, makes full use of the carbon source in the raw water for denitrification, improves the effective utilization rate of carbon source in the raw water, reduces or even eliminates the need for additional carbon source, and at the same time strengthens biological phosphorus removal, achieving "multiple uses of one carbon".

[0022] 4. This invention adopts a dual control mechanism of "feedforward prediction (deep learning) + feedback fine-tuning (PID with dead zone)," which not only ensures rapid response to dynamic operating conditions but also eliminates steady-state errors, ensuring that the ammonia nitrogen in the effluent consistently meets the standards, has strong resistance to shock loads, and enhances the control capability of effluent indicators and the stability of process operation.

[0023] 5. This invention constructs a learning-based intelligent control system. The deep learning model can continuously adapt to long-term changes in parameters such as water quality, water temperature, and sludge concentration through active learning and incremental updates, enabling the control strategy to continuously evolve and possess adaptive and continuous optimization capabilities, thus achieving a leap from "automatic control" to "intelligent optimization".

[0024] 6. This invention actively responds to the national development strategy of "improving quality and efficiency", promotes the green, low-carbon and intelligent transformation of the sewage treatment industry, and provides a replicable and scalable technical solution for the intelligent upgrading of existing sewage treatment plants, with significant economic, environmental and social benefits. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the control system composition in Application Example 1 of the present invention; Figure 2 This invention is SAD n A schematic diagram of the neural network architecture of the model relating to aeration intensity; Figure 3 This is a graph showing the change in aeration rate with influent ammonia nitrogen concentration and influent flow rate in Application Example 1 of the present invention. Figure 4 This is an example of the water intake data diagram of the present invention (Example 1). Figure 5This is a graph showing the monthly average variation trends of carbon source addition ratio, air-to-water ratio, and total nitrogen and total phosphorus in the effluent of the biological tank in the blank and pilot production lines of the wastewater treatment plant in Example 1 of the present invention. Figure 6 This is the water inlet data diagram of Example 2 of the application of the present invention; Figure 7 This is a graph showing the monthly average variation trends of carbon source addition ratio, air-to-water ratio, and total nitrogen and total phosphorus in the effluent of the biological tank in the blank and pilot production lines of the wastewater treatment plant, as described in Example 2 of the application of this invention. Detailed Implementation

[0026] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings in the embodiments of the present invention.

[0027] This invention proposes a dual-loop control method for aerobic space-aeration based on the nitrification process in wastewater treatment. Its core principle is that, unlike the traditional indirect aeration control mode that targets dissolved oxygen, it directly targets the nitrification process as the ultimate goal. That is, ammonia nitrogen removal is used as the fundamental basis for adjusting the aeration rate and aerobic space. By constructing two decoupled yet synergistic control loops of "on-demand aeration" and "on-demand aerobic space", the wastewater treatment process can be precisely controlled and operate efficiently and stably.

[0028] In the "on-demand aeration" loop, this invention proposes a "nitrification ratio gas demand" (SAD). n The core evaluation indicator is the actual amount of air consumed to remove a unit mass of ammonia nitrogen. By establishing a mass balance model that tracks microscopic water masses to a macroscopic sliding window, the SAD (Sarcophaga delta-ammonia) index was determined. n Real-time, accurate calculations are performed; based on this, feedforward control integrates deep learning algorithms to construct multi-dimensional features (water temperature, sludge concentration, load trends, etc.) to an ideal SAD. n The predictive model accurately predicts the future aeration volume required, achieving a leap from "historical average" to "operating condition prediction." At the same time, the introduction of PID feedback control with dead zone corrects minor deviations that exceed the target range, forming a highly efficient collaborative mechanism of "feedforward-led and feedback-fine-tuned" with the feedforward. This loop is directly driven by ammonia nitrogen load, aerating on demand, fundamentally eliminating over-aeration caused by crude dissolved oxygen control, and achieving a dynamic supply and demand balance of aeration volume.

[0029] In the "on-demand aerobic space" loop, this invention breaks through the traditional operational concept of fixed functional zones in biological treatment ponds, proposing to dynamically adjust the aerobic zone space according to the real-time nitrification process; its key lies in revealing the specific gas demand for nitrification (SAD). n The aeration intensity exhibits an inherent "U-shaped" curve pattern, indicating the existence of an "economical aeration intensity" that optimizes energy efficiency. By constructing a neural network model embedding physical priors (U-shaped convexity constraints), the system can accurately solve for the optimal SAD (Savings-Aeration Intensity) under any operating condition.n The minimum economic aeration intensity is determined; then, combined with the target aeration volume obtained from the "on-demand aeration" loop, the optimal aerobic zone volume is calculated; by comparing with the current actual aerobic zone volume, the system intelligently switches the aeration state of the aerobic corridor in a step-by-step manner, under the premise of ensuring process stability (such as setting adjustment dead zones and time locks), converting redundant aerobic tank volume into anoxic zones, thereby increasing the denitrification reaction time.

[0030] The two loops achieve deep decoupling and synergy through the common goal of "nitrification process." The aerobic space loop operates on an hourly scale, adjusting the tank volume to ensure the aeration system always operates near its optimal energy efficiency point. The aeration rate loop operates on a minute scale, precisely matching real-time load changes through feedforward prediction and feedback correction. Ultimately, this invention achieves in-situ, zero-delay intelligent control of the biological nitrification process. While ensuring stable compliance with effluent ammonia nitrogen standards, it maximizes oxygen transfer efficiency and raw water carbon source utilization, significantly reducing aeration power consumption and external carbon source costs, and simultaneously improving total nitrogen removal efficiency and biological phosphorus removal effect. This provides a novel intelligent technical solution for energy saving, consumption reduction, quality improvement, and efficiency enhancement in wastewater treatment plants.

[0031] Example 1 I. Aeration on demand, dynamic supply and demand balance 1. Nitrification ratio gas demand (SAD) n ) In the biological wastewater treatment stage, under certain environmental conditions, the ratio of aeration volume to ammonia nitrogen removal volume obtained by a single small water mass during its complete flow through the aeration zone, i.e., the actual air consumption for removing a unit of ammonia nitrogen, is defined as the Specific Air Demand for Nitrification (SAD). n ).

[0032] 1.1 Instantaneous ratio under ideal steady-state conditions Under steady-state conditions, the total aeration flow rate G (Nm³) in the aeration zone 3 / h)(Nm 3 This refers to the gas volume converted to standard conditions and the influent flow rate Q (m³) in the aeration zone. 3 / h) remains constant.

[0033] Assumption: The volume of the small water mass is V(m) 3 ); ammonia nitrogen concentration at the inlet of the aeration zone N in (g / m 3 ); ammonia nitrogen concentration at the outlet of the aeration zone (N) out (g / m 3 ).

[0034] Within time T (T is the time it takes for the small water mass to travel from the inlet to the outlet of the aeration zone, T=V), 曝气区 / Q, V 曝气区 (The effective volume of the aeration zone) All of this air is ultimately "carried" or comes into contact with the water flowing through it (whether dissolved or not).

[0035] If, during the flow of small water masses through the aeration zone, aeration is uniform and each water mass undergoes the same process, then: The amount of air obtained by the small water mass: (Nm) 3 ) Reduction of ammonia nitrogen in small water masses: (g) Actual air consumption to remove a unit of ammonia nitrogen: (Nm) 3 / kgN) 1.2 Exact Integrals under Time-Varying Conditions Uniform aeration intensity g: Assuming that the aeration intensity (air supply rate per unit volume) is the same at any location within the aeration zone, and equal to the total aeration flow rate G divided by the effective volume V of the aeration zone. 曝气区 That is, g = G / V 曝气区 [(Nm 3 / h) / m 3 At time t, g(t) = G(t) / V(t); this assumption holds approximately when the aerators are uniformly arranged, simplifying the calculation and facilitating engineering applications.

[0036] Plug flow pattern: Macroscopically, the water mass moves in a piston flow manner without longitudinal mixing. Its velocity depends only on the inlet flow rate of the mixed liquor Q(t) of the aeration zone and the cross-sectional area of ​​the tank A (V=A×L, where L is the length of the tank), i.e., ν(t)=Q(t) / A; the distance the water mass moves in time dt is ν(t)dt=(Q(t) / A)dt.

[0037] At any time t (starting from when the water mass enters the aeration zone), the aeration intensity at the location of the water mass is g(t) = G(t) / V(t); since the volume of the water mass is ΔV, the amount of air it "contacts" within a small time dt is: (Nm) 3 ) When the aeration rate G(t) and the inlet mixed liquor flow rate Q(t) of the aeration zone change with time, the calculation of the amount of air obtained by the small water mass needs to be integrated along the trajectory of the small water mass: The amount of air obtained by the small water mass: (Nm)3 ) Where V(t) is the effective volume of the aeration zone (usually constant), and the flow time T is determined by the inlet flow rate of the mixed liquor Q(t) of the aeration zone. In practice, numerical methods are commonly used to solve it.

[0038] Reduction of ammonia nitrogen in small water masses: (g) At this time N out It is the measured value at the moment when the water mass leaves the outlet of the aeration zone, which includes the entire nitrification reaction process.

[0039] Actual air consumption to remove a unit of ammonia nitrogen: (Nm) 3 / kgN) 1.3. Accumulation of sliding windows from micro to macro Since it is impossible to track each water mass individually, the actual control system uses a sliding window mass balance, accumulating the total aeration volume consumed by all water masses within the window with the total ammonia nitrogen removal to obtain the macroscopic SAD. n Calculation formula: (Nm) 3 / kgN) w is the sliding window time, and HT is the time it takes for the mixture to pass through the aeration zone; when the system is in steady state (G, Q, N), w is the sliding window time, and HT is the time it takes for the mixture to pass through the aeration zone. in N out When constant, the two are strictly equal from microscopic to macroscopic levels. Under fluctuating conditions, the window average SAD... n It is a flow-weighted average of instantaneous ratios, which provides greater control stability.

[0040] 2. Real-time aeration volume prediction (feedforward) Use the SAD key in the history window directly. n As a feedforward benchmark, it essentially assumes the SAD (Superior Ability Level) for the next 5-15 minutes. n This is exactly the same as the previous historical window. This assumption holds true when operating conditions are stable, but drastic changes in operating conditions (sudden increase in influent ammonia nitrogen concentration, introduction of inhibitory substances into the influent, sudden changes in sludge concentration, etc.) can lead to SAD. n Rapid changes can cause significant deviations in the calculation of feedforward aeration volume, thereby increasing the burden on feedback regulation and even causing fluctuations in ammonia nitrogen in the effluent.

[0041] 2.1 Deep Learning Algorithms The core of deep learning algorithms is to integrate SAD. n Shifting from "historical averages" to "predictions under current conditions" to forecast future SAD. n,optimal (t), SAD n,optimal(t) Under the current operating condition X(t), the ideal nitrification ratio required to make the effluent ammonia nitrogen reach the target ammonia nitrogen is achieved.

[0042] Deep learning models enable feedforward aeration rate predictions to more closely approximate actual needs, thus improving the accuracy of feedforward predictions. The task of a deep learning model is to learn the mapping relationship f: X(t) → SAD from real-time measurable features X(t). n,optimal (t).

[0043] Deep learning correction and direct use of the nearest window SAD n The essential difference is:

[0044] 2.2 Deep Learning Models (1) Construct training labels (real SAD) n,actual ) From the historical database, time windows where effluent ammonia nitrogen consistently reached the target value and operating conditions were stable were selected, and the actual SAD within these windows was calculated. n , denoted as SAD n,actual Since ammonia nitrogen consistently reaches the target value within this window, SAD can be considered stable. n,actual ≈SAD n,optimal .

[0045] (2) Feature extraction For each qualifying window, extract the measurable feature X for the window period (or the average within the window): historical SAD. n Information such as influent flow rate, biological tank water temperature, ammonia nitrogen concentration in the influent and effluent of the aerobic zone, volume of the aerobic zone, MLVSS, DO, and time.

[0046] (3) Model training Training model f: X→SAD n,optimal This makes the value of f(X) closer to SAD. n,optimal This enables the model to predict SAD under various operating condition combinations. n,optimal .

[0047] (4) Output predicted aeration volume Real-time acquisition of biological system characteristics X(t) at the current moment, model output SAD n,optimal Substituting (X(t)) into the feedforward formula: (Nm) 3 / h) Among them, G a,d (t) represents the predicted aeration rate; Q in (t) represents the inlet flow rate of the mixed liquor in the aeration zone; N in (t) represents the ammonia nitrogen concentration at the inlet of the aeration zone; N tarThe target ammonia nitrogen concentration at the outlet of the aeration zone.

[0048] 3. Engineering Constraints and Modifications (1) Minimum aeration protection When the influent load is extremely low, the algorithm may provide a lower air volume, but the aeration tank still needs to maintain the minimum aeration volume G required for the minimum stirring intensity. a,min (Usually, take 15% to 20% of the maximum air volume, or according to the minimum air volume required to ensure the sludge remains suspended on site).

[0049] (Nm) 3 / h) (2) Maximum aeration rate limit Blower (G) f ), aeration pipes (G) p ), Aeration head (G) s The capabilities of [the system] are limited and an upper limit needs to be set: (Nm) 3 / h) 4. PID feedback correction (feedback) 4.1 PID Controller Algorithm (Incremental) (Nm) 3 / h) (Nm) 3 / h) in: e(k) = N tar N out (k) (mg / L); Δt: Sampling period; U(k): Feedback correction amount ΔQ a (Nm) 3 / h); K p K i K d Parameters need to be adjusted.

[0050] Ultimate target aeration rate: (Nm) 3 / h) Parameter tuning guide:

[0051] 4.2 Engineering constraints that must be added to PID control (1) Dead zone setting Introducing a dead zone in PID control:

[0052]

[0053] Where δ is the dead zone width (typical value 0.2~1.0 mg / L).

[0054] Provided the deep learning model is accurate, the feedforward is sufficient to control ammonia nitrogen within the dead zone, i.e., when the effluent ammonia nitrogen N... out N near the target value tar When the deviation fluctuates within ±δ, the PID output is zero, and aeration is maintained solely by the feedforward. The PID only intervenes to regulate when the deviation exceeds the dead zone, and the larger the deviation, the stronger the regulation. Online ammonia nitrogen instruments have measurement noise, and minor fluctuations should not trigger blower regulation. Setting a dead zone can avoid frequent actions, reduce unnecessary aeration adjustments, and maintain the stability of the aeration system.

[0055] Three operating ranges of PID feedback ① Interval A: Within the dead zone ( ) PID state: Output remains 0 (or integral term is frozen); Aeration settings: (Nm) 3 / h) The control effect relies entirely on the feedforward.

[0056] ②Interval B: Outside the dead zone but within the warning zone ( ) PID state: Proportional + Integral (with derivative) functioning normally; Aeration settings: (Nm) 3 / h) The feedforward method is the primary approach, with feedback as a secondary method, gradually pulling ammonia nitrogen back into the dead zone.

[0057] ③ Interval C: Exceeds the warning value ( ) PID status: Temporarily increase the proportional gain or switch to pure proportional control for a fast response; Aeration settings: (Nm) 3 / h) Feedback-driven operation ensures that the effluent does not exceed the standard; at the same time, the operating condition is recorded for subsequent feedforward model correction.

[0058] Feedforward handles 80% to 90% of the adjustment, while feedback is only used to eliminate residual bias and model error; when SAD n During timed updates, the feedforward model adapts, and the long-term feedback output should approach zero.

[0059] 4.3 The Collaborative Mechanism of PID and Deep Learning Correction The final aeration setting is: (Nm) 3 / h) Predicting future SAD under current operating conditions using a deep learning model n ; This refers to the deviation after dead zone processing.

[0060] Deep learning models are not static; when the PID feedback outputs a non-zero correction value for a long period, it indicates a systematic bias in the feedforward model, and online fine-tuning of the model should be triggered. When the PID output is stable and non-zero and the effluent ammonia nitrogen is at the dead zone boundary, the current operating condition (g,x) is compared with the calculated ideal SAD. n Pairing samples with new ones is used to periodically train the deep neural network, gradually bringing the feedforward closer to the real system. As the running time increases, the frequency and amplitude of PID intervention gradually decrease, and the system becomes increasingly reliant on the feedforward, achieving learning-based control.

[0061] II. On-demand oxygenation and optimal space utilization Under the premise of consistently meeting the effluent ammonia nitrogen standards, the volume of the aerobic zone should always be kept within the economically efficient range under the current operating conditions. The aerobic zone volume should not be too large, leading to excessively low aeration intensity and decreased oxygen transfer efficiency, nor too small, resulting in excessively high aeration intensity, wasted energy, or insufficient mixing. The optimal aerobic zone volume is calculated based on the real-time nitrification gas demand and the economical aeration intensity range of the aeration system; then compared with the current actual aerobic zone volume, the optimal volume is approximated through step-by-step adjustments (turning on / off the aeration channels).

[0062] 1. Aeration intensity and SAD n Association Model 1.1 Economic Aeration Intensity SAD (Saturated Gas Requirement for Nitrification) n Nm 3 The ammonia nitrogen removal rate ( / kgN) is defined as the amount of aeration required to remove a unit mass of ammonia nitrogen, and is a core indicator for evaluating the energy efficiency of the nitrification process.

[0063] Aeration intensity (g, Nm) 3 / m 3 / h) is defined as the air supply per cubic meter of aerobic tank volume per hour, and is a process parameter that can be directly controlled by the aeration system.

[0064] Under fixed operating conditions (water temperature, sludge concentration, influent load, etc.), insufficient aeration (too low g) limits the nitrification rate, while SAD... nWhen the oxygen transfer efficiency (OTE) decreases due to excessive aeration (excessive g), SAD decreases. n Elevated SAD n The aeration intensity g exhibits a U-shaped trend of first decreasing and then increasing, indicating the existence of a unique economic aeration intensity g. opt Make SAD n Reach the global minimum.

[0065] Given the characteristic vector X of the operating condition, the economic aeration intensity g opt (X) is the gas demand SAD required to achieve the nitrification ratio. n Minimum aeration intensity value achieved:

[0066] in, f(g,X): is SAD n The prediction function, representing the nitrification ratio gas requirement when the aeration intensity is g and the operating condition is X, is a strictly convex function with respect to g (second-order partial derivative). 2 f / g 2 (≥α>0), ensuring a globally unique minimum point.

[0067] [g min ,g max The feasible range of aeration intensity (determined by equipment capacity and minimum stirring intensity).

[0068] 1.2, SAD n The principle of U-shaped curve correlation with aeration intensity like Figure 2 As shown, the accuracy of the U-shaped curve is affected by the high-dimensional nonlinear interaction of various operating variables (water temperature, MLVSS, influent ammonia nitrogen load, etc.), and cannot be accurately described by a fixed mechanism formula. Traditional methods use hierarchical discretization (such as dividing by water temperature and load, and fitting a quadratic curve independently for each load), which has inherent defects such as boundary jumps and coarse extreme value estimation.

[0069] This invention proposes to directly learn SAD from aeration intensity and operating condition characteristics using a universal approximator in a continuous space of a neural network (DNN). n The continuous mapping is achieved by embedding physical priors (U-shaped convexity constraints) to ensure the global convexity of the response surface, thereby enabling stable and accurate solution of the economic aeration intensity under arbitrary operating conditions, and utilizing data-driven, adaptive updates.

[0070] 1.3 Association Model (1) Input and output Input: Concatenated vector [g,x], dimension 1+d.

[0071] Output: scalar SAD n Predicted value.

[0072] (2) Network Architecture

[0073] The output layer employs a linear mapping for continuous value regression. Convolutional layers progressively increase the number of channels, extracting local coupling features between g and X through local receptive fields, and completing multi-level representation in a higher-dimensional feature space. BatchNorm stabilizes feature distribution, accelerates convergence, and improves training stability. Dropout uses a strategy of moderate adjustment from shallow to deep to suppress overfitting and preserve effective expressive power. Global average pooling compresses convolutional features and reduces parameter size. The output layer employs a linear mapping to achieve SAD (Simplified Average Array) regression. n Continuous value regression prediction.

[0074] (3) Loss Function Integrating physical priors ①Basic loss: Mean squared error

[0075] To ensure that the model's prediction error on historical data is as small as possible, it learns the true mapping relationship in the data.

[0076] ② Physical regularization (U-shaped convexity constraint)

[0077] in: α > 0: Preset lower bound of curvature (typical value 0.01~0.1), force second derivative ≥ α, ensure SAD n The -g curve is U-shaped (i.e., a convex function), thus it has a unique minimum point; Through automatic differential calculation; λ1: Regularization strength (typical value 0.05~0.2).

[0078] ③ Boundary gradient constraints

[0079] The penalty is that the first derivative at the boundary is non-zero, which guides the minimum value to fall within the feasible region and avoids extreme values.

[0080] Total loss:

[0081] By adjusting the hyperparameters λ1, λ2, and α, an optimal balance can be achieved between fitting accuracy and physical plausibility.

[0082] (4) Association Model Let the historical compliance data set be:

[0083] The average aeration intensity (Nm³) during the i-th stable operating window 3 / m 3 / h); : The corresponding operating condition feature vector (water temperature, MLVSS, influent ammonia nitrogen load, etc.); y i =SAD n,i : Nitrification ratio gas demand (Nm³) calculated within the window 3 / kgN).

[0084] Constructing continuous functions (parameter θ ), so that:

[0085] And for any X, f θ (g,X) in [g min ,g max Strictly convex on the upper surface, for any real-time operating condition Quickly solve g opt ( ).

[0086] (5) Continuous model updates ① Proactive learning (covering sparse operating conditions) Regularly analyze the coverage density of historical data across various operating conditions; For low-density areas, a stepped aeration intensity disturbance experiment is automatically performed during low-load periods to collect (g, x, SAD) data under compliant operating conditions. n )sample.

[0087] ② Incremental learning Newly qualified samples will be added to the training set daily; Full retraining once a week to ensure long-term stability.

[0088] ③ Rollback mechanism When the model outputs g opt If the result exceeds the physical range three times consecutively (e.g., <g_min or >g_max), or the deviation from the stratified fitting result is greater than 20%, the system will automatically switch to the stratified fitting method and issue an alarm.

[0089] This invention proposes a method for continuous fitting and solving of economic aeration intensity based on neural networks. This is achieved by constructing a SAD (Synchronous Aeration Detection) system. nBy employing a high-dimensional continuous mapping of f(g,X) and utilizing second-order regularization to force U-shaped convexity, this method achieves real-time, sub-step-size accurate estimation of economic aeration intensity under arbitrary operating conditions while ensuring physical rationality; it completely eliminates boundary jumps in hierarchical fitting; and through active learning and incremental updates, the model adapts in the long term.

[0090] 2. Algorithm for dynamic spatial adjustment of aerobic zone Using the aeration rate prediction curve for the next 8 hours Nitrification Ratio Gas Demand Prediction Curve Calculate the volume demand curve for the aerobic zone in the next 8 hours. And while ensuring that the effluent meets the standards, the space of the aerobic zone is adjusted slowly and in a step-by-step manner to make the actual aeration intensity approach the economic value, thereby maximizing the aeration efficiency.

[0091] 2.1 Required volume based on aeration intensity If the total aeration volume Q is... a With economic aeration intensity g opt If the distribution is uniform, the required volume of the aerobic zone is:

[0092] Economic aeration intensity g opt The neural network model predicts the nitrification ratio gas demand (SAD) for the next 8 hours. n Minimum gas supply per unit tank volume (Nm³) 3 / m 3 / h).

[0093] Target aeration rate Q a The total aeration volume (Nm³) required to achieve effluent quality standards in the next 8 hours, predicted by a neural network model. 3 / h).

[0094] 2.2 Required volume based on nitration kinetics To ensure sufficient nitrification reaction time, minimum hydraulic retention time or sludge loading constraints must be met: or

[0095] Q in Inflow rate (m³) 3 / h); N int : Ammonia nitrogen in influent (mg / L); N target Target effluent ammonia nitrogen (mg / L); AUR: Ammonia nitrogen utilization rate (mgN / gVSS·h), which can be determined by regression from historical data or online. MLVSS: Volatile suspended solids concentration in the mixture (g / L); HRT min Minimum hydraulic retention time, set according to water temperature and mud age.

[0096] 2.3 Final Required Volume

[0097] Where V min This is the absolute minimum volume of the aerobic zone in the process (e.g., 20% of the total tank volume).

[0098] 3. Adjust decision-making logic (step-by-step) To minimize the frequency of aeration zone space adjustments while ensuring process stability, adjustment decisions must be shifted from immediate feedback to proactive planning. For example, based on predictions of nitrification gas demand over the next 8 hours, the suitability of the current aerobic zone volume for future periods should be assessed. Adjustments should only be implemented when there is a long-term imbalance between the predicted load and the current volume, extending the adjustment cycle to over 6 hours. To avoid frequent adjustments, a dead zone + time lock strategy is employed.

[0099] 3.1 Input Current effective volume V of the aerobic zone current (m) 3 ); Calculate the required volume V of the aerobic zone req (m) 3 ); Switchable aeration corridors / cells: Volume V per cell cell Current state (aerobic / hypoxic); Time t since the last spatial adjustment last (h).

[0100] 3.2 Deviation Calculation

[0101]

[0102] 3.3 Triggering conditions (simultaneously met) Time lock + deviation exceeding limit: within the next T (e.g., 6 hours), ,recommend =0.1-0.15 (10%-15%) Safety verification: The adjusted aerobic zone volume is not less than V. min The volume of the hypoxic zone shall not be less than that of the required hypoxic zone.

[0103] 3.4 Adjustment of Implementation like V > 0 (current volume is too large): Select the corresponding aeration unit that can be turned off from the end of the aerobic zone and switch it to hypoxia.

[0104] like V < 0 (current volume is too small): Select the appropriate aeration unit to turn on from the end of the aerobic zone and switch it to aerobic mode.

[0105] Update after each adjustment and record the adjustment time.

[0106] This algorithm uses the economic aeration intensity g opt With real-time target aeration rate Q a The ratio directly provides the required aerobic zone volume and integrates a biodynamic kernel to ensure process safety. A dead-zone + time lock + step-by-step switching strategy significantly reduces adjustment frequency, forming minute-to-hour-level synergy with PID control, enabling the aeration system to operate near its "optimal efficiency point."

[0107] Application Example 1 A wastewater treatment plant in Zhengzhou, Henan Province, with a designed capacity of 350,000 tons / day, is operating at full capacity. Its influent data is as follows: Figure 4 As shown. The biological treatment tank of this wastewater treatment plant uses the Paton process. Two production lines were selected as experimental blank controls, and the other two production lines were used as experimental samples.

[0108] On June 8, 2025, the pilot production line began applying the technology of this invention (Add), and the control system is as follows: Figure 1 As shown, the application results are as follows: Figure 3 , 5 As shown: It can be seen that the method of this invention can effectively ensure the stability of effluent ammonia nitrogen. The blank and experimental production lines share the same influent distribution well, with an annual average total nitrogen content of 55.3 mg / L. The total nitrogen content in the secondary effluent of the blank and experimental production lines is 14.1 mg / L and 7.8 mg / L, respectively, with removal rates of 74.4% and 85.9%, respectively. The experimental line shows an improvement of 11.5% compared to the blank production line. The total phosphorus content in the influent is 4.82 mg / L, while the total phosphorus content in the secondary effluent of the blank and experimental production lines is 0.18 mg / L and 0.13 mg / L, respectively. The total phosphorus removal rate exceeds 96% in both cases, and the experimental line shows a reduction of 27.8% in effluent total phosphorus compared to the blank production line. The external carbon source dosage per ton of water in the blank and experimental production lines is 38.6 mg / L and 1.5 mg / L, respectively, with the experimental line showing a reduction of 96% compared to the blank production line. The air-to-water ratio in the blank and experimental production lines is 4.36 and 3.64, respectively, with the experimental line showing a reduction of 16.4% compared to the blank production line. The annual cost savings for carbon sources are approximately 4.74 million yuan.

[0109] Application Example 2 A wastewater treatment plant in Xinmi, Henan Province, with a designed daily treatment capacity of 30,000 tons, is operating at full capacity. Its influent data is as follows: Figure 6As shown. The biological treatment tank of this wastewater treatment plant uses the Paton process, with a total of 2 production lines. One production line was selected as a control, and the other production line was used as an experiment.

[0110] The technology of this invention was first applied to a pilot production line on August 3, 2025. The application results are as follows: Figure 7 As shown: The blank and experimental production lines share the same inlet water distribution well. The average total nitrogen in the inlet is 68.5 mg / L. The total nitrogen in the secondary effluent of the blank and experimental production lines is 10.4 mg / L and 4.6 mg / L, respectively, with removal rates of 84.8% and 93.3%, respectively. The experimental line shows an 8.5% improvement over the blank. The total phosphorus inlet is 5.4 mg / L. The total phosphorus in the secondary effluent of the blank and experimental production lines is 0.51 mg / L and 0.13 mg / L, respectively, with removal rates of 90.5% and 97.6%, respectively. The experimental line shows a 7.1% improvement over the blank. The external carbon source dosage per ton of water in the blank and experimental production lines is 31 mg / L and 0 mg / L, respectively, with the experimental line showing a 100% reduction compared to the blank. The air-to-water ratio in the blank and experimental production lines is 6.67 and 5.27, respectively, with the experimental line showing a 20.6% reduction compared to the blank. The annual cost savings from carbon source reduction are approximately 340,000 yuan.

[0111] This invention is not limited to the preferred embodiments described above. Anyone can derive other forms of products under the guidance of this invention. However, regardless of any changes made in their shape or structure, any technical solution that is the same as or similar to this application falls within the protection scope of this invention.

Claims

1. A wastewater treatment aerobic space-aeration dual-loop control method based on the nitrification process, characterized in that: This includes on-demand aeration circuits and on-demand aerobic space circuits; The on-demand aeration loop acquires the influent flow rate, influent ammonia nitrogen concentration, and effluent ammonia nitrogen concentration in the aeration zone in real time, and calculates the nitrification ratio air demand (SAD). n SAD n The actual amount of air consumed to remove a unit mass of ammonia nitrogen; based on SAD n Based on the target ammonia nitrogen removal load, calculate and predict the aeration rate, and combine feedback correction to accurately calculate the target aeration rate and control the aeration equipment to supply air as needed; Establish SAD on demand aerobic space loop n The correlation model with aeration intensity g, where aeration intensity g is the ratio of the total aeration volume to the effective volume of the aeration zone, is used to solve for SAD. n Minimum economic aeration intensity g opt Based on g opt Calculate the optimal aerobic zone volume V based on the target aeration rate. req V req Compared with the current actual aerobic zone volume V current Compare and dynamically adjust the volume of the aerobic zone based on the comparison results; The on-demand aeration loop and the on-demand aerobic space loop are decoupled based on the nitrification process and controlled collaboratively.

2. The wastewater treatment aerobic space-aeration dual-loop control method based on nitrification process according to claim 1, characterized in that: Based on an accurate integral model for tracking microscopic water clumps, a sliding window accumulation is performed from the microscopic to the macroscopic level to obtain the SAD. n The calculation formula is as follows: (Nm) 3 / kgN); Where G is the total aeration flow rate, Q is the influent flow rate of the aeration zone, and N is the total aeration flow rate. in The ammonia nitrogen concentration in the influent of the aeration zone, N out t represents the ammonia nitrogen concentration in the effluent from the aeration zone, w represents the sliding window time, and HT represents the time it takes for the mixed liquor to pass through the aeration zone.

3. The wastewater treatment aerobic space-aeration dual-loop control method based on nitrification process according to claim 1, characterized in that: The feedforward control algorithm is used to calculate the predicted aeration rate. The calculation formula for the feedforward control algorithm is as follows: (Nm) 3 / h) Among them, G a,req To predict aeration volume, Q in N is the influent flow rate to the aeration zone. in The ammonia nitrogen concentration in the influent of the aeration zone, N tar The target effluent ammonia nitrogen concentration for the aeration zone.

4. The wastewater treatment aerobic space-aeration dual-loop control method based on nitrification process according to claim 3, characterized in that: The SAD mentioned n The optimal nitrification ratio gas demand (SAD) predicted by the deep learning model n,optimal (t); Deep learning models are used to learn the mapping relationship f: X(t) → SAD from real-time measurable operating condition features X(t). n,optimal (X(t)); Operating characteristic X(t) includes historical SAD n One or more of the following: influent flow rate, biological tank water temperature, ammonia nitrogen concentration in the influent and effluent of the aerobic zone, volume of the aerobic zone, MLVSS, DO, and time; The formula for calculating the predicted aeration rate using a deep learning model is as follows: (Nm) 3 / h).

5. The wastewater treatment aerobic space-aeration dual-loop control method based on nitrification process according to claim 1, characterized in that: The feedback correction employs a PID control algorithm with dead zone; The aforementioned PID control algorithm with dead zone has a deviation e dead (t) is defined as: in, This indicates a deviation in the control of ammonia nitrogen in the effluent. This is the preset dead zone width; Target aeration rate G a,set (k) Aeration rate G predicted by feedforward a,d (k) is formed by superimposing the PID feedback correction amount u(k): G a,set (k)=G a,d (k)+u(k).

6. The wastewater treatment aerobic space-aeration dual-loop control method based on nitrification process according to claim 1, characterized in that: In the aforementioned on-demand aerobic spatial loop, a neural network model is constructed to learn the gas demand for nitrification ratio (SAD) from aeration intensity g and operating condition characteristics X. n continuous mapping f θ (g,X); By introducing a physical prior regularization term into the loss function, it is ensured that for any operating condition X, the mapping f θ (g,X) is a strictly convex function within the feasible region of aeration intensity, i.e., it satisfies the U-shaped curve characteristic; by solving for the global minimum point of the strictly convex function, the economic aeration intensity g under the current operating conditions can be obtained. opt (X).

7. The wastewater treatment aerobic space-aeration dual-loop control method based on nitrification process according to claim 1, characterized in that, The loss function includes: Mean squared error term used to ensure the accuracy of model predictions: ； Physical regularization terms used to enforce U-shaped convexity constraints: ； in, To predetermine the lower bound of curvature, The regularization strength; Boundary gradient constraints used to guide the minimum value to fall within the feasible region: 。 8. The wastewater treatment aerobic space-aeration dual-loop control method based on nitrification process according to claim 1, characterized in that, The optimal aerobic zone volume The calculation formula is: in, The volume requirement for the aerobic zone is based on the aeration intensity; Target aeration rate; Minimum volume requirement based on nitration kinetics school core; This is the absolute minimum volume of the aerobic zone in the process.

9. The wastewater treatment aerobic space-aeration dual-loop control method based on nitrification process according to claim 1, characterized in that, The aforementioned dynamic adjustment of the aerobic zone size employs a step-by-step adjustment strategy, specifically including: Configure and adjust the dead zone ratio and time lock (TIock); When the deviation ratio Exceeding the dead zone ratio An adjustment is triggered when the time since the last adjustment exceeds the time lock Tlock. like > Turn off the corresponding number of aeration units at the end of the aerobic zone to switch it to the anoxic zone; if Start the corresponding number of aeration units at the end of the aerobic zone to switch it to the aerobic zone.

10. The control system of the method according to any one of claims 1-9, characterized in that: It includes a data acquisition module, an on-demand aeration control module, an on-demand aerobic space control module, and a core processing module; The data acquisition module is used to collect process parameters in real time, including influent flow rate in the aeration zone, ammonia nitrogen concentration in the influent of the aeration zone, ammonia nitrogen concentration in the effluent of the aeration zone, water temperature in the biological tank, MLVSS, and DO. The on-demand aeration control module includes SAD n The calculation unit, feedforward prediction unit, and feedback correction unit are used to execute the on-demand aeration loop, generate the target aeration volume, and send it to the aeration actuator. The on-demand aerobic space control module includes an association model construction unit, an economic aeration intensity solution unit, an optimal aerobic zone volume calculation unit, and a space adjustment decision unit, which are used to execute the on-demand aerobic space loop, generate space adjustment instructions, and send them to the aeration valve or control terminal. The core processing module is used to run deep learning algorithms, store historical data, and coordinate the collaborative work of the on-demand aeration control module and the on-demand aerobic space control module.

Images