A method and system for the coordinated supply of a boiler feed water pH conditioning agent

Abstract

The application relates to a boiler feed water pH adjusting agent synergistic supply method and system, and relates to the technical field of pH adjustment, and comprises the following steps: collecting real-time operation data of a pH adjusting unit after a deaerator; calculating water body buffer capacity according to the real-time operation data, determining a theoretical basic agent amount of acid-base agents; predicting future effluent pH values and scaling risk indexes through an integrated learning model; constructing a multi-objective optimization function with pH deviation, pH fluctuation, scaling risk, minimum agent cost and maximum system stable operation time; and adopting a model predictive control algorithm to combine the theoretical basic agent amount and the predicted values to rollingly solve the optimal synergistic dosing amount of acid agents, alkali agents and scale inhibitors. The scheme provided by the application can fuse buffer capacity mechanism constraints and data-driven prediction, realize accurate synergistic supply of acid-base and scale inhibitors, improve pH control precision and stability, and inhibit scaling risks.

Description

Technical Field

[0001] This application relates to the field of pH adjustment technology, and in particular to a method and system for the coordinated supply of pH adjustment agents to boiler feedwater. Background Technology

[0002] In boiler feedwater treatment systems, the pH adjustment unit after the deaerator is used to control the pH value of the feedwater to avoid acidic corrosion and alkaline scaling. In actual operation, hydraulic parameters such as feedwater quality, flow rate, and pressure often fluctuate simultaneously, resulting in significant time-varying characteristics in the demand for acid and alkali reagents and the buffer response strength.

[0003] The pH adjustment unit after the deaerator has a high outlet water temperature, resulting in large pH measurement drift and slow response at high temperatures. Low feedwater pH can lead to acid corrosion of economizer and boiler heating surfaces, while high pH can promote the deposition of carbonates, silicates, and phosphates. The addition of acid and alkali chemicals is coupled with the supply of scale inhibitors. Simply adjusting the pH may exacerbate the risk of scaling and increase the cost of chemicals.

[0004] In existing technologies, pH adjustment often employs threshold-based start / stop or fixed-ratio dosing methods. These methods cannot proactively identify pH lag changes caused by sudden flow rate fluctuations, nor can they actively adjust scale inhibitors or boiler operating cycles when pH meets standards but scaling risk increases. Some technologies use theoretical dosage calculations based on buffer capacity mechanism models, providing theoretical dosing references, but they cannot fully characterize the higher-order coupling effects of variables such as phosphate, silica, and hardness on scaling risk. Another type of technology uses purely data-driven models to fit historical operating data. While this has strong nonlinear modeling capabilities, it is prone to significant prediction bias when the influent water quality deviates significantly from the training sample distribution, and it lacks physical interpretability, making it difficult to guarantee control reliability under extreme operating conditions.

[0005] Therefore, there is an urgent need for a method that integrates buffer capacity mechanism constraints with data-driven prediction to achieve synergistic supply of acids, bases and scale inhibitors, while simultaneously optimizing pH accuracy, stability, scaling risk, reagent cost and runtime. Summary of the Invention

[0006] To address or partially address the problems existing in related technologies, this application provides a method and system for the coordinated supply of pH adjustment agents for boiler feedwater, aiming to solve the technical problems of pH adjustment lag, inaccurate agent control, and conflict between pH standard attainment and scale inhibition in existing pH adjustment units after deaerators.

[0007] This application provides a method for the coordinated supply of pH adjusting agents to boiler feedwater, including: Collect real-time operating data of the pH adjustment unit after the boiler feedwater deaerator. The operating data should include at least the pH value, hardness index, alkalinity and conductivity of the influent and effluent. The water body's buffer capacity is dynamically calculated based on the collected real-time operational data, and the theoretical basic dosage of acid and alkali agents is calculated based on the buffer capacity. Among them, a theoretical basic dosage greater than the standard value is used for high alkalinity water bodies, and a theoretical basic dosage less than the standard value is used for low alkalinity water bodies. Based on operational data, the effluent pH value and scaling risk index are predicted in the future control time domain using a pre-trained ensemble learning model. A multi-objective optimization function is constructed with the comprehensive optimization objectives of minimizing pH target value deviation, pH fluctuation, scaling risk, reagent cost, and maximizing system stable operating time. Based on real-time operating data, theoretical dosage, predicted effluent pH, and predicted scaling risk, a model predictive control algorithm is used to continuously solve for the optimal control sequence in the future control time domain. The optimal control sequence includes at least the dosage of acid agent, alkali agent, and scale inhibitor. Based on the optimal control sequence obtained by joint solution, the dosing pump is controlled to perform dosing operation, and the synergistic supply of acid, alkali and scale inhibitor is adjusted synchronously.

[0008] Optionally, the water body's buffer capacity can be dynamically calculated based on the collected real-time operational data, including: Calculate the buffer capacity of the current water body based on alkalinity, total hardness, conductivity, and temperature: In the formula, For alkalinity, Here, E is the total hardness, EC is the electrical conductivity, and T is the temperature. For reference temperature, , , , These are coefficients calibrated through offline experiments; Set high and low alkalinity thresholds; when the alkalinity exceeds the high alkalinity threshold, the buffer capacity... An increase in alkalinity indicates an enhanced buffering capacity of the water body to acidic and alkaline reagents; when alkalinity is below the low alkalinity threshold, the buffering capacity... A decrease in the value indicates a weakening of the water body's buffering capacity against acid and alkali agents.

[0009] Optionally, the theoretical basis for calculating the dosage of acid-base agents based on buffer capacity includes: Obtain the deviation between the current effluent pH and the target pH. and the current water supply flow rate Q; Calculate the theoretical drug dose per unit flow rate based on the buffer capacity. : In the formula, k is the equivalent coefficient of the drug, which is determined according to the type and concentration of the drug used; Apply a nonlinear correction to the theoretical drug dosage, when When the deviation exceeds the upper limit, an acceleration factor is introduced. To accelerate the pH response; when When the deviation is less than the lower limit and the water body's buffer capacity is lower than the low capacity threshold, a deceleration factor is introduced. To prevent pH overshoot, the corrected theoretical dosage .

[0010] Optionally, a multi-objective optimization function is constructed with the comprehensive optimization objectives of minimizing pH target value deviation, pH fluctuation, scaling risk, reagent cost minimization, and maximizing system stable operating time, including: Define the optimization objective: Minimize the deviation of pH from the target value, minimize pH fluctuation, minimize scaling risk, minimize reagent costs, and maximize the stable operating time of the system; Minimize the squared deviation between the current effluent pH and the target pH, denoted as , The pH value of the effluent. The target pH value; Minimize pH fluctuation coefficient The pH fluctuation coefficient is defined as the coefficient of variation of pH value, that is, the ratio of pH standard deviation to pH mean, and is used to evaluate the stability of pH control. Minimize the scale risk index The scaling risk index is calculated by weighting at least five parameters from calcium hardness, magnesium hardness, silicon ions, phosphate, conductivity, temperature, pH and water pressure fluctuations. Minimize the total cost of the drug per unit time. The total cost of the reagents is the sum of the products of the amount of acid, alkali and scale inhibitor used and their corresponding unit prices; Maximize system stable runtime Stable operating time requires that the pH is within the target range, the scaling risk index is below the safety threshold, the water supply pressure fluctuation is below the preset limit, and the dosing pump does not frequently start and stop. The five optimization objectives are integrated into a multi-objective optimization function: In the formula, , , , These are adaptive weighting coefficients; When the scaling risk index When the risk threshold is exceeded, it will automatically increase. And correspondingly reduce other weights to prioritize the suppression of scaling risk.

[0011] Optionally, a model predictive control algorithm is used to solve for the optimal control sequence in the future control time domain in a rolling manner, including: In each control cycle, the current effluent pH value, water quality parameters, and feedwater flow rate are obtained, and the predicted effluent pH value and scaling risk index for the future control time domain are received. The multi-objective optimization function is used as the cost function for rolling optimization, and the control time domain length is set. and prediction time domain length , ≤ ; Set constraints: Chemical dosage variation rate constraint: The variation range of acid, alkali or scale inhibitor dosage within a single control cycle shall not exceed the preset maximum variation rate threshold; Equipment safety boundary constraints: The pump operating frequency shall not exceed the minimum and maximum values ​​allowed by the frequency converter, and the valve operating frequency shall not exceed the mechanical life threshold. Hydraulic hysteresis compensation constraint: The cross-correlation method is used to estimate the time delay between the change in influent water quality and the pH response of effluent. The delay is used as the offset of the input feature to align the adjustment time of the controller with the effluent response time. Within the prediction time domain, the optimal control sequence that minimizes the cost function and maximizes the stable operating time is solved. The optimal control sequence includes the acid dosage, alkali dosage, scale inhibitor dosage, and operating frequency of the variable frequency dosing pump within the future control time domain. Only the control quantity of the first control cycle in the optimal control sequence obtained by the solution is output to the actuator, and the solution steps are repeated in the next control cycle to achieve rolling optimization.

[0012] Optional hydraulic lag compensation constraints include: Obtain the influent pH change sequence Response sequence of effluent pH The influent pH change sequence is the detected value of influent pH within a historical time window, and the effluent pH response sequence is the detected value of effluent pH within the same time window. The degree of time delay correlation between two sequences is calculated using the cross-correlation function: For discrete sampled data: In the formula, To delay the step size, The cross-correlation value is N, where N is the number of sampling points; Determine the cross-correlation value Maximum delay step : Will The corresponding time length is used as the hydraulic lag compensation delay. In the input feature configuration of the model predictive control algorithm, the influent pH, water quality parameters and dosing command at the current moment are time-shifted according to the delay amount, so that the controller's adjustment time corresponds to the effluent response time. When system traffic changes, the delay is dynamically adjusted based on the ratio of the current traffic to the rated traffic. In the formula, To calibrate the flow rate The measured delay amount, This represents the current water supply flow rate.

[0013] Optional, the constraint on the rate of change of drug dosage includes: Obtain the dosage of acid, alkali and scale inhibitor in the previous control cycle; Calculate the absolute changes in acid, alkali and scale inhibitors in the current control cycle compared to the previous control cycle; The rate of change is constrained, and the single adjustment range of acid, alkali, and scale inhibitor does not exceed the preset maximum change threshold. Among them, the maximum change threshold of each acid, alkali, and scale inhibitor is independently calibrated based on the hydraulic impact tolerance limit of the pipeline, the response time of the pH sensor, and the mechanical characteristics of the reagent pump. When the single-step adjustment of any agent in the optimal control sequence obtained by the model predictive control algorithm exceeds the corresponding maximum change threshold, the agent dosage is restricted. Based on the dosage of the agent in the previous control cycle, the increase or decrease shall not exceed the threshold. The rate of change constraint is enforced during the rolling optimization process in each control cycle to prevent hydraulic shock in pipelines, brief misreading of pH probes, and chemical stress damage to boiler pipelines caused by sudden changes in reagent flow.

[0014] Optionally, in the process of using the model predictive control algorithm to solve for the optimal control sequence in the future control time domain, when the pH control objective conflicts with the scaling risk suppression objective, the scaling risk index is incorporated as a hard constraint into the multi-objective optimization function. In each control cycle of the model predictive control algorithm, it is first determined whether the scaling risk index in the current prediction period exceeds the preset safety threshold. If it does, the scaling risk index is reduced to below the safety threshold as the first optimization objective, and the control sequence that minimizes the pH deviation is solved under this constraint. Only when the scaling risk index is already below the safety threshold is the pH deviation minimized as the main optimization objective.

[0015] A second aspect of this application provides a boiler feedwater pH adjustment agent co-supply system, comprising: The multimodal sensing module is used to collect real-time operating data of the pH adjustment unit after the boiler feedwater deaerator. The operating data includes influent and effluent pH, total hardness, calcium hardness, magnesium hardness, alkalinity, phosphate, silicon ions, conductivity, temperature, oxidation-reduction potential (ORP), turbidity, suspended solids (SS), iron and manganese ions, particle count, fouling density index (SDI), flow rate, and deaerator outlet pressure. The derived index calculation module is used to calculate derived indices based on operating data. These derived indices include pH adjustment accuracy, pH fluctuation coefficient, reagent dosage, scaling inhibition rate, feedwater pressure fluctuation, boiler operating cycle, flow fluctuation coefficient, and system stable operating time. The buffer capacity mechanism module estimates the water body's response intensity to acid and alkali agents based on alkalinity, hardness, conductivity, and temperature, calculates the water body's buffer capacity, and determines the theoretical basis dosage of acid and alkali agents. The prediction module uses an ensemble learning model with hyperparameters optimized by a Bayesian optimization algorithm to predict the effluent pH value and scaling risk index in the future control time domain. The scaling risk assessment module calculates the scaling risk index by comprehensively considering calcium hardness, magnesium hardness, silicon ions, phosphate, conductivity, temperature, pH, and water pressure fluctuations. The multi-objective MPC decision module takes pH deviation, pH fluctuation, reagent cost, scaling risk and stable operating time as optimization objectives. It combines theoretical dosage, predicted effluent pH value, predicted scaling risk value and scaling risk index to solve the acid and alkali reagent dosage, scale inhibitor dosage and variable frequency dosing pump operating frequency in a rolling manner. The execution control module is used to generate corresponding control commands based on the calculated dosage, and send the control commands to the corresponding strain frequency dosing pump to synchronously adjust the synergistic supply of acid, alkali and scale inhibitor; The stability feedback module is used to perform online correction of model parameters based on stable runtime and error feedback.

[0016] The technical solution provided in this application may include the following beneficial effects: By dynamically calculating the water body's buffer capacity, the theoretical dosage of chemicals is determined based on the differences between high and low alkalinity water bodies. An integrated learning model is used to predict future pH and scaling risks. A multi-objective function is constructed to simultaneously optimize pH deviation, pH fluctuation, scaling risk, chemical cost, and stable operating time. The model predicts and controls the synergistic dosage of acid, alkali, and scale inhibitors, improving the response speed and stability of pH adjustment, avoiding over- or under-dosing, effectively suppressing scaling risk while ensuring pH compliance, reducing chemical costs, and extending the system's stable operating time. This solves the problems of pH control lag, inaccurate chemical control, and conflicting objectives.

[0017] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0018] The above and other objects, features and advantages of this application will become more apparent from the more detailed description of exemplary embodiments thereof in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments thereof.

[0019] Figure 1 This is a schematic flowchart illustrating the method for co-supplying pH adjustment agents for boiler feedwater according to an embodiment of this application. Detailed Implementation

[0020] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to make this application more thorough and complete, and to fully convey the scope of this application to those skilled in the art.

[0021] The technical solutions of the embodiments of this application are described in detail below with reference to the accompanying drawings.

[0022] In some embodiments, a boiler feedwater pH adjusting agent co-supply system includes: The module includes a multimodal perception module, a derived index calculation module, a buffer capacity mechanism module, a prediction module, a scaling risk assessment module, a multi-objective MPC decision-making module, an execution control module, and a stability feedback module.

[0023] The multimodal sensing module is used to collect real-time operating data from the pH adjustment unit after the boiler feedwater deaerator. This data includes influent and effluent pH, total hardness, calcium hardness, magnesium hardness, alkalinity, phosphate, silicon ions, conductivity, temperature, oxidation-reduction potential (ORP), turbidity, suspended solids (SS), iron and manganese ions, particle count, fouling density index (SDI), flow rate, and deaerator outlet pressure. All sensors collect data at a preset sampling frequency, and the data is stored in a real-time database after quality control.

[0024] The derived index calculation module is used to calculate derived indices based on operational data. These derived indices include: pH adjustment accuracy, pH fluctuation coefficient, reagent dosage, scaling inhibition rate, feedwater pressure fluctuation, boiler operating cycle, flow fluctuation coefficient, and system stable operating time. Specifically, pH adjustment accuracy is the absolute deviation between the effluent pH and the target pH; pH fluctuation coefficient is the coefficient of variation of pH values, i.e., the ratio of the pH standard deviation to the pH mean; reagent dosage is the dosage of acid, alkali, and scale inhibitor per unit time; scaling inhibition rate is the decrease in the scaling risk index per unit time; feedwater pressure fluctuation is the pressure change or rate of change; boiler operating cycle is the cumulative operating time after this cleaning or maintenance; flow fluctuation coefficient is the coefficient of variation of the flow rate; and system stable operating time is the continuous operating time that simultaneously meets the following conditions: effluent pH is within the preset target range, the scaling risk index is below the safety threshold, feedwater pressure fluctuation is below the preset limit, and the dosing pump has not frequently started or stopped.

[0025] The buffer capacity mechanism module is used to estimate the response intensity of water to acid and alkali agents based on alkalinity, hardness, conductivity, and temperature, calculate the buffer capacity of the water body, and determine the theoretical basis dosage of acid and alkali agents. Specifically, it calculates the current buffer capacity of the water body based on alkalinity, total hardness, conductivity, and temperature. : (1) In the formula, For alkalinity, Here, E is the total hardness, EC is the electrical conductivity, and T is the temperature. For reference temperature, , , , These are coefficients calibrated through offline experiments; The buffer capacity mechanism module also sets high and low alkalinity thresholds: when the alkalinity is greater than the high alkalinity threshold, the water body is judged to have a high buffer capacity. If the alkalinity value increases, a theoretically based dosage greater than the baseline value is used; when the alkalinity is less than the low alkalinity threshold, the water body is judged to have low buffer capacity. As the value decreases, a theoretically based drug dosage lower than the baseline value is used. Subsequently, based on the pH deviation... Given the water flow rate Q, calculate the theoretical basis dosage of the reagent. , where k is the equivalent dosage coefficient. The theoretical dosage is output to the multi-objective MPC decision module as the initial reference value for the model predictive control algorithm.

[0026] The prediction module is used to predict the effluent pH and scaling risk index within the future control time domain. This module is equipped with an ensemble learning model optimized by hyperparameters. It uses current and historical influent and effluent pH, alkalinity, hardness, conductivity, temperature, flow rate, buffer capacity β, and current reagent dosage as input features to predict the effluent pH and scaling risk values ​​within the future control time domain. The ensemble learning model is trained offline using a rolling time-series validation method and updated periodically online.

[0027] The scaling risk assessment module is used to calculate a scaling risk index by comprehensively considering calcium hardness, magnesium hardness, silicon ion concentration, phosphate concentration, conductivity, temperature, pH, and feedwater pressure fluctuations. Specifically, the above parameters are normalized separately and then weighted and summed. When the pH control target conflicts with the scaling risk suppression target, the scaling risk index serves as a hard constraint term in the multi-objective optimization, prioritizing ensuring that the scaling risk index is below a safe threshold before optimizing the pH deviation.

[0028] The multi-objective MPC decision module is used to optimize pH deviation, pH fluctuation, reagent cost, scaling risk, and stable operating time. It combines theoretical dosage, predicted effluent pH, and predicted scaling risk to continuously solve for the dosage of acid and alkali reagents, the dosage of scale inhibitor, and the operating frequency of the variable frequency dosing pump. In each control cycle, this module performs the following operations: acquires the current state variables, and sets the control time domain length. and prediction time domain length , ≤ The multi-objective function, which includes the square of pH deviation, pH fluctuation coefficient, scaling risk index, minimization of reagent cost, and maximization of stable operating time, is used as the cost function for rolling optimization. Under the conditions of satisfying the constraints of reagent dosage change rate, equipment safety boundary constraints, and hydraulic lag compensation constraints, the optimal control sequence that minimizes the cost function is solved, and only the control quantity of the first control cycle is output.

[0029] The execution control module generates corresponding control commands based on the calculated dosages of acid and alkali agents, scale inhibitors, and the operating frequency of the variable frequency dosing pump. These commands are then sent to the corresponding variable frequency dosing pump to synchronously adjust the combined supply of acid, alkali, and scale inhibitors. A safety check is performed before the commands are issued; if the control quantity output by the model exceeds the equipment's safety boundary, it is automatically truncated to the safety boundary value, and the reason for truncation is recorded.

[0030] The stability feedback module is used to perform online calibration of model parameters based on stable runtime and error feedback. A candidate model mechanism is employed: candidate models are first trained using the latest data. The online model is only replaced if the candidate model outperforms the current model in terms of prediction error, false negative rate, and false positive rate within the validation window; otherwise, only the statistical baseline and safety threshold are updated. When operators manually intervene, this module marks the data during the intervention period as manual intervention samples and isolates them from the training set. Only after confirming stable improvement from the intervention are these samples incorporated into the model learning as expert strategy samples.

[0031] In some embodiments, corresponding to the aforementioned application function implementation system embodiments, this application also provides a method for the coordinated supply of boiler feedwater pH adjusting agents and corresponding embodiments.

[0032] See Figure 1 A method for co-supplying a pH adjusting agent for boiler feedwater, comprising: S101. Collect real-time operating data of the pH adjustment unit after the boiler feedwater deaerator; Specifically, real-time operational data of the pH adjustment unit after the boiler feedwater deaerator is collected. This data includes at least the influent and effluent pH values, total hardness, calcium hardness, magnesium hardness, alkalinity, conductivity, temperature, and feedwater flow rate. The collected data undergoes the following preprocessing: Timestamp resampling unifies all instrument data to a preset baseline timestamp, and linear interpolation is used to align delayed data.

[0033] For missing value imputation, linear interpolation of adjacent valid points is used for short-term missing values; for long-term missing values ​​or sensor freezes, they are marked as untrainable samples and the previous valid value is used to maintain them in real-time control.

[0034] Outlier identification employs a two-layer logic, combining the 3σ rule, the absolute deviation of the moving median, and process boundaries. The first layer identifies statistical anomalies, using the 3σ rule and the absolute deviation of the moving median in parallel or in series to identify outliers that do not conform to statistical laws. The second layer identifies process consistency. For data deemed "normal" by the first layer, further consistency checks are performed using process boundaries to identify samples that are statistically normal but have illogical process logic. For example, a flow sensor value remaining constant while a pressure sensor value abruptly changes by more than a preset percentage is marked as "hydraulic inconsistency"; a pH sensor value remaining constant while an acid / alkali agent flow rate suddenly increases is marked as "dosing and pH mismatch"; and significant changes in effluent indicators within a time frame shorter than the hydraulic residence time are marked as "time logic error." Identified outlier samples are not directly deleted but are sent to an anomaly buffer queue, where the safety control logic replaces them with conservative boundary values.

[0035] Process boundary normalization involves normalizing each valid feature based on the physical upper and lower limits of the process parameters, while preserving the original engineering quantities, normalized quantities, rates of change, and moving window statistics. This enables the model to identify both absolute exceedances and early risks that have not yet exceeded limits but have abnormal slopes.

[0036] S102. Calculate the water body buffer capacity dynamically based on the collected real-time operation data, and calculate the theoretical basis dosage of acid and alkali agents based on the buffer capacity. Specifically, the buffering capacity of the current water body is calculated based on alkalinity, total hardness, conductivity, and temperature. (1) In the formula, Alkalinity; Total hardness; EC is electrical conductivity; T is temperature; For reference temperature, 25℃ is usually taken; , , , The coefficients were determined through offline experiments. The method involved taking typical water samples in the laboratory or on-site, changing the alkalinity, hardness, conductivity, and temperature, recording the pH changes caused by a unit dosage of the reagent, and then using multivariate nonlinear regression fitting to obtain the coefficients.

[0037] Based on boiler feedwater quality standards or historical operating data, high and low alkalinity thresholds are determined. When alkalinity exceeds the high alkalinity threshold, the buffer capacity... An increase in alkalinity indicates a stronger buffering capacity of the water body to acid and alkali reagents, requiring an increase in the dosage; when alkalinity is below the low alkalinity threshold, the buffering capacity... A decrease in the value indicates a weakening of the water body's buffering capacity against acid and alkali agents, requiring a reduction in the dosage to prevent over-flushing.

[0038] Obtain the deviation between the current effluent pH and the target pH. and the current water supply flow rate Q; Calculate the theoretical drug dose per unit flow rate based on the buffer capacity. : (2) In the formula, k is the reagent equivalent coefficient, which is calibrated according to the type and concentration of the reagent used. For alkali reagents, the k value is determined by experimental measurement of the pH increase of a unit concentration of alkali reagent; for acid reagents, the k value is determined by experimental measurement of the pH decrease of a unit concentration of acid reagent; for different concentration batches, the k value can be corrected online.

[0039] A nonlinear correction is applied to the theoretical drug dosage to avoid slow response under large deviations and overshoot under low buffering under small deviations. Upper deviation threshold, lower deviation threshold, and low capacity threshold are set. When the deviation exceeds the upper limit, an acceleration factor is introduced. To accelerate the pH response; when When the deviation is less than the lower limit and the water body's buffer capacity is lower than the low capacity threshold, a deceleration factor is introduced. To prevent pH overshoot, the corrected theoretical dosage .

[0040] S103. Based on the operational data, predict the effluent pH value and scaling risk index in the future control time domain using a pre-trained ensemble learning model. Specifically, the pre-trained ensemble learning model adopts a Stacking ensemble learning model based on TPE Bayesian optimization; the Stacking ensemble learning model adopts a two-layer structure: Base learner layer: Four heterogeneous models are used simultaneously: XGBoost, LightGBM, CatBoost, and Extremely Random Tree. Each base learner independently receives input features, including influent and effluent pH, total hardness, calcium hardness, magnesium hardness, alkalinity, phosphate, silica, conductivity, temperature, redox potential, turbidity, suspended solids, iron and manganese ions, flow rate, deaerator outlet pressure and buffer capacity, and current reagent dosage within the current and past preset time windows. Each base learner outputs a residual prediction component or a direct prediction component, which is the predicted deviation between the mechanistic prediction baseline and the actual value.

[0041] Second-level meta-learner layer: Using a temporal convolutional network (TCN) or a Transformer Encoder, the meta-learner receives the predicted components output by the four base learners as input feature sequences. It uses causal convolution, dilated convolution, or self-attention mechanisms to capture long-term dependencies across boiler operation cycles, sedimentation reaction cycles, and pH adjustment cycles, and outputs the final predicted values ​​for effluent pH and scaling risk.

[0042] TPE Bayesian hyperparameter optimization The optimal hyperparameter combination is automatically searched using the Tree-structured Parzen Estimator (TPE) algorithm. The specific process is as follows: The validation error y of each candidate hyperparameter x is used as the observation value. Candidate hyperparameter x includes the learning rate, tree depth, number of leaf nodes, regularization coefficient, sampling ratio, time window length, and prediction step size. The set of parameters with validation errors below the quantile threshold y is denoted as the good set, and the rest as the bad set. Kernel density estimation is used to model the probability density of each set. and The desired improvement in EI is defined as: (3) The TPE algorithm prioritizes selecting [the appropriate algorithm] in the next search round. The optimal hyperparameter combination is obtained after a preset number of iterations over the largest parameter region.

[0043] Model training employs a rolling time-series validation approach, prohibiting random sample shuffling. Historical data is divided into training, validation, and testing intervals in chronological order. Future data remains invisible within each validation window to prevent information leakage during the continuous water treatment process. The training set is used to train the Stacking model, the validation set is used for error evaluation in TPE optimization, and the test set is used to evaluate the generalization performance of the final model.

[0044] Online prediction, during the online operation phase, uses real-time operational data from the current moment and past time windows as input, and continuously outputs predicted effluent pH values ​​for the future control time domain. and scale risk prediction value , where h is the prediction step size. Short prediction step sizes are used for real-time control, while long prediction step sizes are used for maintenance planning and early warning.

[0045] S104. Construct a multi-objective optimization function with the comprehensive optimization objectives of minimizing pH target value deviation, pH fluctuation, scaling risk, reagent cost, and maximizing system stable operating time; Specifically, the optimization objectives are defined as minimizing the deviation of pH from the target value, minimizing pH fluctuation, minimizing scaling risk, minimizing reagent costs, and maximizing the stable operating time of the system. Minimize the squared deviation between the current effluent pH and the target pH. , The pH value of the effluent. The target pH value is used to penalize deviations from the target pH value, and is expressed in a squared form to increase the penalty weight for deviations.

[0046] Minimize pH fluctuation coefficient The pH fluctuation coefficient is defined as the coefficient of variation of pH value, that is, the ratio of pH standard deviation to pH mean. It is used to evaluate the stability of pH control and suppress frequent oscillations.

[0047] Minimize the scale risk index The scaling risk index is calculated by weighting at least five parameters from calcium hardness, magnesium hardness, silicon ion concentration, phosphate concentration, conductivity, temperature, pH, and feedwater pressure fluctuations. Specifically, each participating parameter is normalized, then multiplied by its corresponding weighting coefficient, and the results are summed, with the sum of all weighting coefficients equal to 1. When the pH control target conflicts with the scaling risk suppression target, the scaling risk index is used as a hard constraint term in the multi-objective optimization, prioritizing the protection of pH. Below the safety threshold.

[0048] Minimize the total cost of the drug per unit time. , = Acid dosage × Acid unit price + Alkali dosage × Alkali unit price + Scale inhibitor dosage × Scale inhibitor unit price. The unit price can be updated in real time according to the purchase price.

[0049] Maximize system stable runtime The stable operating time requirement is to simultaneously meet the following conditions: pH within the target range, scaling risk index below the safety threshold, water supply pressure fluctuation below the preset limit, and no frequent start-stop of the dosing pump. This target serves as a reward, prompting the controller to prioritize strategies that maintain stable operation.

[0050] The five optimization objectives are integrated into a multi-objective optimization function: (4) In the formula, , , , For adaptive weighting coefficients, satisfying The initial weights can be set according to the importance of the process; when the scaling risk index... When the preset risk threshold is exceeded, the system will automatically... Increase, and , , Reduced proportionally to prioritize the suppression of scaling risk, when Once the weights are restored to below the safe level, they are returned to their initial values.

[0051] In the rolling solution of the actual model predictive control algorithm, the system stable operating time can be maximized. It is added to the cost function as a negative reward term, i.e. ,in This is a preset positive coefficient.

[0052] S105. Based on real-time operating data, theoretical dosage, predicted effluent pH, and predicted scaling risk, a model predictive control algorithm is used to continuously solve for the optimal control sequence in the future control time domain; the optimal control sequence includes at least the dosage of acid agent, alkali agent, and scale inhibitor. Specifically, within each control cycle, the following sub-steps are executed: State Acquisition and Prediction: Acquire the effluent pH value, water quality parameters, and feedwater flow rate at the current time t, and receive the future prediction time domain length output by the prediction module. pH prediction within the step and scale risk prediction value ,in .

[0053] Set optimization variables: Set the control time domain length ,satisfy ≤ Optimize variables for the future. Acid dosage within each control cycle Alkali dosage Dosage of scale inhibitor and the operating frequency of the variable frequency dosing pump , .

[0054] Constructing the cost function: The multi-objective optimization function is converted into a discrete form, which serves as the rolling optimization cost function for the rolling solution of the Model Predictive Control (MPC) algorithm. (5) In the formula, Calculated based on the pH prediction sequence within the prediction window according to , , calculate.

[0055] Apply constraints: Constraints on the rate of change of drug dosage: , This represents the maximum permissible variation of the acid in a single cycle; similarly, it applies to the alkali. , has scale inhibitor ; The values ​​are calibrated independently based on the pipeline hydraulic shock tolerance limit, pH sensor response time, and the mechanical characteristics of the reagent pump. For example, and It can be taken as 10% to 20% of the pump's rated capacity. A range of 5% to 10% is acceptable.

[0056] Equipment safety boundary constraints: pump operating frequency ≤ ≤ The dosage of each agent shall not exceed the maximum supply capacity of the pump; the valve operating frequency shall not exceed the mechanical life threshold.

[0057] The hydraulic lag compensation constraint uses a cross-correlation method to estimate the time delay between changes in influent water quality and effluent pH response. The delay is used as the offset of the input feature to align the controller's adjustment time with the effluent response time.

[0058] Hydraulic lag compensation is also one of the constraints: Obtain the influent pH change sequence within the historical time window. and effluent pH response sequence The influent pH change sequence is the detected value of influent pH within a historical time window, and the effluent pH response sequence is the detected value of effluent pH within the same time window. The degree of time delay correlation between two sequences is calculated using the cross-correlation function: (6) For discrete sampled data: (7) In the formula, This is the delay step size, ranging from 0 to the preset maximum delay. The cross-correlation value is N, where N is the number of sampling points; Determine the cross-correlation value Maximum delay step : (8) Will The corresponding time length is used as the hydraulic lag compensation delay. .

[0059] In the input feature configuration of the Model Predictive Control (MPC) algorithm, the current influent pH, water quality parameters, and dosing instructions are configured according to... Perform time offset alignment so that the controller's adjustment time corresponds to the water output response time.

[0060] When system traffic changes, based on the current traffic... With the rated flow rate The ratio dynamically adjusts the delay amount: .

[0061] Solving the optimization problem: Employ appropriate optimization algorithms, such as sequential quadratic programming, gradient descent, or enumeration combined with constraint handling, to solve the problem while satisfying the aforementioned constraints. Minimum preferred control sequence .

[0062] Rolling execution: Only the control quantity of the first control cycle in the optimal control sequence obtained by solving is output to the execution control module. In the next control cycle, the state and prediction are updated, and the above steps are repeated to achieve rolling optimization.

[0063] Conflict handling occurs when the pH control target conflicts with the scaling risk inhibition target. If the scale risk index exceeds the safety threshold, it is incorporated into the optimization as a hard constraint. Specifically, the scale risk index is incorporated into the multi-objective optimization function as a hard constraint. In each control cycle of the model predictive control algorithm, it is first determined whether the scale risk index in the current prediction period exceeds the preset safety threshold. If it does, the first optimization objective is to reduce the scale risk index to below the safety threshold. Under this constraint, the control sequence that minimizes the pH deviation is solved. Only when the scale risk index is already below the safety threshold is the minimization of the pH deviation taken as the main optimization objective.

[0064] S106. Based on the optimal control sequence obtained by joint solution, control the dosing pump to perform dosing operation and synchronously adjust the synergistic supply of acid, alkali and scale inhibitor. Specifically, the execution control module, based on the first set of control quantities in the optimal control sequence obtained by the Model Predictive Control (MPC) rolling solution, sends commands to the variable frequency dosing pump via the industrial communication protocol to synchronously adjust the dosage of acid, alkali, and scale inhibitor. Before execution, the execution control module performs safety checks on all outputs. Equipment constraint verification: If the pump frequency output by the model exceeds the minimum or maximum value allowed by the frequency converter, the value will be automatically truncated to the boundary value; if the amount of reagent added exceeds the upper limit of the pump's mechanical capacity, the upper limit value will be taken.

[0065] Process constraint verification: If the output causes the predicted pH value of the effluent to exceed the dangerous range, or the predicted value of the scaling risk index to exceed the emergency threshold, the command will be rejected and the system will switch to conservative control mode.

[0066] Drug concentration drift correction: When the effective concentration changes due to batch replacement or long-term storage of the drug, the system estimates the effective concentration correction coefficient online by comparing the theoretical pH response with the actual pH response, and calculates the control quantity according to the coefficient, so that the same pump frequency produces a consistent control effect under different drug batches.

[0067] All limiting, rejection, or correction events are logged on the host computer, and operational suggestions are generated for maintenance personnel to refer to.

[0068] S107. Based on the system's stable operating time and error feedback, a candidate model mechanism is used to perform online correction of the model parameters; Stable runtime monitoring: Real-time monitoring of changes in stable runtime. If the pH fluctuation coefficient exceeds the preset upper limit for multiple consecutive control cycles, the MPC control step size is automatically reduced, and the weight of the buffer capacity model in the initial reference value is increased to enhance mechanistic constraints. If the stable runtime increases and the unit drug consumption decreases, the current parameter set is marked as an excellent sample and sent to the TPE optimization library for subsequent hyperparameter initialization or constraint range adjustment.

[0069] Candidate Model Mechanism: The system maintains the current online model and candidate models. After a preset time interval or when sufficient new data has been accumulated, the candidate model is retrained using the latest operational data from the past period. Training data includes normal operating conditions, load adjustments, water quality fluctuations, and cleaning / recovery scenarios. After training, three performance metrics are evaluated within a reserved validation window: The root mean square error (RMSE) of the prediction is used to measure the accuracy of pH prediction. The over-limit underreporting rate is the proportion of samples where the actual pH or scaling risk index exceeds the safety threshold but the model does not issue a warning. False alarm rate: the proportion of samples that receive a warning from the model but do not actually exceed the safety threshold; The candidate model is replaced with the new online model only if all three metrics of the candidate model are better than the current online model; otherwise, the current online model is retained, and only the statistical baseline and the safety boundary in the model predictive control algorithm (MPC) cost function are updated.

[0070] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for co-supplying a pH adjusting agent to boiler feedwater, characterized in that, include: Collect real-time operating data of the pH adjustment unit after the boiler feedwater deaerator. The operating data includes at least the pH value of the influent and effluent, hardness index, alkalinity and conductivity. The water body buffer capacity is dynamically calculated based on the collected real-time operation data, and the theoretical basic dosage of acid and alkali agents is calculated based on the buffer capacity; wherein, for high alkalinity water bodies, a theoretical basic dosage greater than the standard value is used, and for low alkalinity water bodies, a theoretical basic dosage less than the standard value is used. Based on the operational data, the effluent pH value and scaling risk index in the future control time domain are predicted using a pre-trained ensemble learning model. A multi-objective optimization function is constructed with the comprehensive optimization objectives of minimizing pH target value deviation, pH fluctuation, scaling risk, reagent cost, and maximizing system stable operating time. Based on the real-time operating data, theoretical dosage, predicted effluent pH, and predicted scaling risk, a model predictive control algorithm is used to continuously solve for the optimal control sequence in the future control time domain. The optimal control sequence includes at least the dosage of acid agent, alkali agent, and scale inhibitor. Based on the optimal control sequence obtained by joint solution, the dosing pump is controlled to perform dosing operation, and the synergistic supply of acid, alkali and scale inhibitor is adjusted synchronously.

2. The method for co-supplying boiler feedwater pH adjusting agents according to claim 1, characterized in that, The process of dynamically calculating the water body buffer capacity based on collected real-time operational data includes: Calculate the buffer capacity of the current water body based on alkalinity, total hardness, conductivity, and temperature: In the formula, For alkalinity, Here, E is the total hardness, EC is the electrical conductivity, and T is the temperature. For reference temperature, , , , These are coefficients calibrated through offline experiments; Set high and low alkalinity thresholds; when the alkalinity exceeds the high alkalinity threshold, the buffer capacity... An increase in alkalinity indicates an enhanced buffering capacity of the water body to acid and alkali reagents; when alkalinity is below the low alkalinity threshold, the buffering capacity... A decrease in the value indicates a weakening of the water body's buffering capacity against acid and alkali agents.

3. The method for co-supplying boiler feedwater pH adjusting agents according to claim 1, characterized in that, The theoretical basis for calculating the dosage of acid-base agents based on the buffer capacity includes: Obtain the deviation between the current effluent pH and the target pH. and the current water supply flow rate Q; Calculate the theoretical drug dose per unit flow rate based on the buffer capacity. : In the formula, k is the equivalent coefficient of the drug, which is determined according to the type and concentration of the drug used; Apply a nonlinear correction to the theoretical drug dosage, when When the deviation exceeds the upper limit, an acceleration factor is introduced. To accelerate the pH response; when When the deviation is less than the lower limit and the water body's buffer capacity is lower than the low capacity threshold, a deceleration factor is introduced. To prevent pH overshoot, the corrected theoretical dosage .

4. The method for co-supplying boiler feedwater pH adjusting agents according to claim 1, characterized in that, The construction of a multi-objective optimization function with the comprehensive optimization objectives of minimizing pH target value deviation, pH fluctuation, scaling risk, reagent cost minimization, and maximizing system stable operating time includes: Define the optimization objective: Minimize the deviation of pH from the target value, minimize pH fluctuation, minimize scaling risk, minimize reagent costs, and maximize the stable operating time of the system; Minimize the squared deviation between the current effluent pH and the target pH, denoted as , The pH value of the effluent. The target pH value; Minimize pH fluctuation coefficient The pH fluctuation coefficient is defined as the coefficient of variation of pH value, that is, the ratio of pH standard deviation to pH mean, and is used to evaluate the stability of pH control. Minimize the scale risk index The scaling risk index is calculated by weighting at least five parameters from calcium hardness, magnesium hardness, silicon ions, phosphate, conductivity, temperature, pH and water pressure fluctuations. Minimize the total cost of the drug per unit time. The total cost of the reagent is the sum of the products of the amount of acid, alkali and scale inhibitor used and their corresponding unit prices; Maximize system stable runtime Stable operating time requires that the pH is within the target range, the scaling risk index is below the safety threshold, the water supply pressure fluctuation is below the preset limit, and the dosing pump does not frequently start and stop. The five optimization objectives are integrated into a multi-objective optimization function: In the formula, , , , These are adaptive weighting coefficients; When the scaling risk index When the risk threshold is exceeded, it will automatically increase. And correspondingly reduce other weights to prioritize the suppression of scaling risk.

5. The method for co-supplying boiler feedwater pH adjusting agents according to claim 1, characterized in that, The method of using a model predictive control algorithm to solve for the optimal control sequence in the future control time domain includes: In each control cycle, the current effluent pH value, water quality parameters, and feedwater flow rate are obtained, and the predicted effluent pH value and scaling risk index for the future control time domain are received. The multi-objective optimization function is used as the cost function for rolling optimization, and the control time domain length is set. and prediction time domain length , ≤ ; Set constraints: Chemical dosage variation rate constraint: The variation range of acid, alkali or scale inhibitor dosage within a single control cycle shall not exceed the preset maximum variation rate threshold; Equipment safety boundary constraints: The pump operating frequency shall not exceed the minimum and maximum values ​​allowed by the frequency converter, and the valve operating frequency shall not exceed the mechanical life threshold. Hydraulic hysteresis compensation constraint: The cross-correlation method is used to estimate the time delay between the change in influent water quality and the pH response of effluent. The delay is used as the offset of the input feature to align the adjustment time of the controller with the effluent response time. Within the prediction time domain, the optimal control sequence that minimizes the cost function and maximizes the stable operating time is solved. The optimal control sequence includes the acid dosage, alkali dosage, scale inhibitor dosage, and operating frequency of the variable frequency dosing pump within the future control time domain. Only the control quantity of the first control cycle in the optimal control sequence obtained by the solution is output to the actuator, and the solution steps are repeated in the next control cycle to achieve rolling optimization.

6. The method for co-supplying boiler feedwater pH adjusting agents according to claim 5, characterized in that, The hydraulic lag compensation constraint includes: Obtain the influent pH change sequence Response sequence of effluent pH The influent pH change sequence is the detected value of influent pH within a historical time window, and the effluent pH response sequence is the detected value of effluent pH within the same time window. The degree of time delay correlation between two sequences is calculated using the cross-correlation function: For discrete sampled data: In the formula, To delay the step size, The cross-correlation value is N, where N is the number of sampling points; Determine the cross-correlation value Maximum delay step : Will The corresponding time length is used as the hydraulic lag compensation delay. In the input feature configuration of the model predictive control algorithm, the influent pH, water quality parameters and dosing command at the current moment are time-shifted and aligned according to the delay amount, so that the controller's adjustment time corresponds to the effluent response time. When system traffic changes, the delay is dynamically adjusted based on the ratio of the current traffic to the rated traffic. In the formula, To calibrate the flow rate The measured delay amount, This represents the current water supply flow rate.

7. The method for co-supplying boiler feedwater pH adjusting agents according to claim 5, characterized in that, The constraint on the rate of change of drug dosage includes: Obtain the dosage of acid, alkali and scale inhibitor in the previous control cycle; Calculate the absolute changes in acid, alkali and scale inhibitors in the current control cycle compared to the previous control cycle; The rate of change is constrained, and the single adjustment range of acid, alkali, and scale inhibitor does not exceed the preset maximum change threshold. Among them, the maximum change threshold of each acid, alkali, and scale inhibitor is independently calibrated based on the hydraulic impact tolerance limit of the pipeline, the response time of the pH sensor, and the mechanical characteristics of the reagent pump. When the single-step adjustment of any agent in the preferred control sequence obtained by the model predictive control algorithm exceeds the corresponding maximum change threshold, the agent dosage is restricted. Based on the dosage of the agent in the previous control cycle, the increase or decrease shall not exceed the threshold. The rate of change constraint is enforced during the rolling optimization process in each control cycle to prevent hydraulic shocks in pipelines, brief misreads of pH probes, and chemical stress damage to boiler pipelines caused by sudden changes in reagent flow.

8. The method for co-supplying boiler feedwater pH adjusting agents according to claim 5, characterized in that: In the process of using the model predictive control algorithm to solve the optimal control sequence in the future control time domain, when the pH control target conflicts with the scaling risk suppression target, the scaling risk index is incorporated as a hard constraint into the multi-objective optimization function. In each control cycle of the model predictive control algorithm, it is first determined whether the scaling risk index in the current prediction period exceeds the preset safety threshold. If the scale risk index is exceeded, the first optimization objective is to reduce the scale risk index to below the safety threshold. Under this constraint, the control sequence that minimizes the pH deviation is solved. Only when the scale risk index is already below the safety threshold is the pH deviation minimized as the main optimization objective.

9. A boiler feedwater pH adjusting agent co-supply system, used to execute the boiler feedwater pH adjusting agent co-supply method according to any one of claims 1-8, characterized in that, include: The multimodal sensing module is used to collect real-time operating data of the pH adjustment unit after the boiler feedwater deaerator. The operating data includes influent and effluent pH, total hardness, calcium hardness, magnesium hardness, alkalinity, phosphate, silicon ions, conductivity, temperature, redox potential, turbidity, suspended solids, iron and manganese ions, particle count, fouling density index, flow rate, and deaerator outlet pressure. The derived index calculation module is used to calculate derived indices based on the operating data. The derived indices include pH adjustment accuracy, pH fluctuation coefficient, reagent dosage, scaling inhibition rate, feedwater pressure fluctuation, boiler operating cycle, flow fluctuation coefficient, and system stable operating time. The buffer capacity mechanism module estimates the water body's response intensity to acid and alkali agents based on alkalinity, hardness, conductivity, and temperature, calculates the water body's buffer capacity, and determines the theoretical basis dosage of acid and alkali agents. The prediction module uses an ensemble learning model with hyperparameters optimized by a Bayesian optimization algorithm to predict the effluent pH value and scaling risk index in the future control time domain. The scaling risk assessment module calculates the scaling risk index by comprehensively considering calcium hardness, magnesium hardness, silicon ions, phosphate, conductivity, temperature, pH, and water pressure fluctuations. The multi-objective MPC decision module takes pH deviation, pH fluctuation, reagent cost, scaling risk and stable operating time as optimization objectives. It combines theoretical dosage, predicted effluent pH value, predicted scaling risk value and scaling risk index to solve the acid and alkali reagent dosage, scale inhibitor dosage and variable frequency dosing pump operating frequency in a rolling manner. The execution control module is used to generate corresponding control commands based on the calculated dosage, and send the control commands to the corresponding strain frequency dosing pump to synchronously adjust the synergistic supply of acid, alkali and scale inhibitor; The stability feedback module is used to perform online correction of model parameters based on stable runtime and error feedback.

Images