SO2 concentration early prediction and closed-loop control method and system

Abstract

The application discloses a SO2 concentration advanced prediction and closed-loop control method and system, which is applied to a wet flue gas desulfurization system. The method acquires and pre-processes desulfurization operation parameters in real time, calls a mass transfer-reaction coupling mechanism model to obtain a mechanism predicted value of an outlet SO2 concentration, obtains a data compensation item based on a data-driven deviation compensation model, and fuses to generate an advanced predicted value of the outlet SO2 concentration in a future prediction time domain. According to the advanced predicted value, an SO2 concentration outer loop control instruction is generated, a slurry pH set point is corrected, a pH inner loop control instruction is generated, and a slurry flow output and a circulating pump frequency control instruction are output. The scheme can reduce the influence of SO2 feedback control lag, improve the prediction accuracy and the closed-loop control stability.

Description

Technical Field

[0001] This application belongs to the field of flue gas environmental protection control technology for coal-fired power plants, specifically relating to a method and system for predicting and controlling SO2 concentration in advance, and more particularly to a method and system for predicting and controlling SO2 concentration at the outlet based on mechanism prediction and data compensation fusion, applicable to wet desulfurization systems. Background Technology

[0002] Wet flue gas desulfurization (FGD) is the mainstream process for SO2 emission reduction in coal-fired power plants. During operation, wet FGD systems are affected by various factors such as inlet flue gas SO2 concentration, flue gas flow rate, boiler load, slurry pH value, slurry density, slurry supply flow rate, and circulation pump frequency, and typically exhibit characteristics such as large inertia, pure time delay, strong nonlinearity, and strong coupling of multiple variables.

[0003] Existing wet desulfurization control methods mostly use the measured SO2 concentration at the absorber outlet as the feedback quantity, and control the slurry flow rate or the operating status of the circulating pump through PID control to achieve outlet SO2 concentration control. However, since the measured outlet SO2 concentration lags significantly behind the internal reaction process of the desulfurization system, the feedback control based on the measured value is difficult to respond in a timely manner under operating conditions such as load fluctuations, inlet SO2 concentration disturbances, and changes in the operating status of the circulating pump, which can easily cause short-term fluctuations in the outlet SO2 concentration.

[0004] To address these issues, some existing solutions employ mechanistic or data-driven models to predict the desulfurization process. Mechanistic models typically rely on desulfurization reaction kinetics and gas-liquid mass transfer theory, offering some physical interpretation. However, under complex operating conditions such as slurry aging, salt accumulation, uneven flow fields, and equipment characteristic drift, model parameters can easily deviate from actual operating conditions, resulting in insufficient long-term prediction accuracy. Data-driven models can learn the system's dynamic characteristics using operational data, but lack physical constraints, leading to insufficient generalization ability and extrapolation reliability under conditions such as variable loads and pump start-up / shutdown.

[0005] Therefore, it is necessary to provide a method and system for SO2 concentration advance prediction and closed-loop control that can take into account both mechanistic constraints and data compensation capabilities, so as to reduce the impact of outlet SO2 concentration feedback lag on control effect and improve the control stability of wet desulfurization system under complex operating conditions. Summary of the Invention

[0006] The purpose of this application is to provide a method for SO2 concentration advance prediction and closed-loop control, which addresses the technical problems of existing wet desulfurization SO2 concentration control, such as feedback lag, insufficient prediction accuracy of single models, and inadequate adaptability to complex operating conditions.

[0007] To achieve the above objectives, this application adopts the following technical solution: Firstly, this application proposes a method for predicting and controlling SO2 concentration in advance, including: The operating parameters of the wet desulfurization system are acquired in real time, and the operating parameters are preprocessed. The pre-processed operating parameters are input into the pre-established SO2 concentration mechanism model to obtain the mechanism prediction value of the outlet SO2 concentration; Based on the predicted value of the mechanism and the measured value of the outlet SO2 concentration, the dynamic deviation is determined, and the dynamic deviation is input into a pre-established data-driven deviation compensation model to obtain the data compensation term. The mechanism prediction value is fused with the data compensation term to obtain the mixed prediction value of the outlet SO2 concentration at the future prediction time. An outer loop control for SO2 concentration is constructed based on the predicted mixed value of the outlet SO2 concentration, and an inner loop control for pH is constructed based on the pH value of the slurry, so that the outer loop control for SO2 concentration and the inner loop control for pH can work together to generate control commands for the wet desulfurization system.

[0008] Furthermore, the operating parameters include inlet flue gas parameters, slurry state parameters, control execution parameters, and controlled output parameters; The inlet flue gas parameters include inlet SO2 concentration, flue gas flow rate, boiler load, and flue gas temperature; The slurry state parameters include slurry pH value, slurry density, and slurry temperature; The control execution parameters include slurry flow rate and circulation pump frequency; The controlled output parameters include the measured value of SO2 concentration at the absorber outlet.

[0009] Furthermore, the preprocessing of the operating parameters includes at least one of outlier removal, bad point filtering, missing value interpolation, timing alignment, time delay compensation, normalization, measurement point validity judgment, and redundancy verification.

[0010] Furthermore, the SO2 concentration mechanism model is a mass transfer-reaction coupling model established based on desulfurization reaction kinetics and gas-liquid mass transfer theory; The SO2 concentration mechanism model takes at least the inlet SO2 concentration, flue gas flow rate, boiler load, slurry pH value, slurry density, overall mass transfer coefficient and reaction rate constant as inputs or model parameters, and outputs the mechanism prediction value. The overall mass transfer coefficient and the reaction rate constant are determined through field operating condition tests.

[0011] Furthermore, the dynamic deviation is the difference between the measured value of the outlet SO2 concentration and the mechanism prediction value at the corresponding time. The data-driven bias compensation model includes a time series model formed by combining VMD and LSTM, and / or a time series model formed by combining VMD and LSSVM.

[0012] Furthermore, under rapidly changing operating conditions, LSSVM is invoked to learn and predict the dynamic deviation, and under strongly nonlinear long-term operating conditions, LSTM is invoked to learn and predict the dynamic deviation.

[0013] Furthermore, the predicted mixed SO2 concentration at the outlet satisfies: C_hybrid(t+τ)=C_mech(t+τ)+C_data(t+τ); Where C_hybrid(t+τ) is the predicted mixed SO2 concentration at the outlet at the future prediction time, C_mech(t+τ) is the predicted mechanistic concentration at the future prediction time, C_data(t+τ) is the data compensation term at the future prediction time, and τ is the prediction lead time; The prediction lead time τ is adaptively adjusted in real time based on the unit load, the number of circulating pumps in operation, and the liquid level in the absorption tower, and τ is 15s to 60s.

[0014] Furthermore, the outer loop control of SO2 concentration includes: comparing the predicted value of SO2 concentration at the outlet with the set value of SO2 emission, and outputting a circulating pump frequency adjustment command based on the comparison result; The pH inner loop control includes: receiving the control command of the SO2 concentration outer loop control, and dynamically correcting the pH setpoint according to the control command; When generating control commands for the wet desulfurization system, dynamic weight allocation is performed based on SO2 prediction deviation, and multi-objective optimization is carried out on the circulation pump frequency and pH setting range, with SO2 compliance, pH stability and gypsum quality as constraints. Among them, when the SO2 prediction deviation is greater than 10 mg / m³ 3 When SO2 concentration control weight is increased; when SO2 prediction deviation is less than or equal to 10 mg / m³ 3 At that time, reduce the SO2 concentration control weight.

[0015] Secondly, this application proposes an SO2 concentration advance prediction and closed-loop control system, comprising: The data acquisition module is used to acquire the operating parameters of the wet desulfurization system in real time and to preprocess the operating parameters. The mechanism prediction module is used to input the pre-processed operating parameters into the pre-established SO2 concentration mechanism model to obtain the mechanism prediction value of the outlet SO2 concentration. The deviation compensation module is used to determine the dynamic deviation based on the predicted value of the mechanism and the measured value of the outlet SO2 concentration, and input the dynamic deviation into a pre-established data-driven deviation compensation model to obtain the data compensation item. The fusion prediction module is used to fuse the mechanism prediction value with the data compensation item to obtain the mixed prediction value of the outlet SO2 concentration at the future prediction time. The closed-loop control module is used to construct an outer loop control of SO2 concentration based on the predicted mixed value of SO2 concentration at the outlet, and to construct an inner loop control of pH based on the pH value of the slurry, so that the outer loop control of SO2 concentration and the inner loop control of pH work together to generate control commands for the wet desulfurization system.

[0016] Furthermore, the data acquisition module acquires the operating parameters in real time through the DCS / OPC interface; The closed-loop control module includes an outer loop control unit for SO2 concentration and an inner loop control unit for pH. The outer loop control unit for SO2 concentration is used to compare the predicted value of SO2 concentration at the outlet with the set value of SO2 emission, and output a circulating pump frequency adjustment command based on the comparison result. The inner loop control unit for pH is used to receive the control command from the outer loop control unit for SO2 concentration, and dynamically correct the pH set point based on the control command. The system also includes an online optimization module and a safety protection module. The online optimization module is used to feed back the measured values ​​of outlet SO2 concentration and slurry pH to the hybrid model formed by the SO2 concentration mechanism model and the data-driven deviation compensation model, and update the parameters of the data-driven deviation compensation model online. The safety protection module is used to execute at least one of the following: abnormal deviation alarm, model freeze protection, model failure judgment, fault-safe logic, and non-intrusive PID switchback mechanism.

[0017] Compared with the prior art, this application has the following beneficial effects: This application acquires and preprocesses the operating parameters of a wet flue gas desulfurization system in real time. Combined with a pre-constructed mass transfer-reaction coupling mechanism model and a data-driven deviation compensation model, it predicts the SO2 concentration at the absorber outlet in advance. The prediction results are then used for outer-loop SO2 concentration control and inner-loop slurry pH control, enabling control actions to intervene before the actual SO2 concentration exceeds the standard or fluctuates significantly, reducing the lag effect of traditional control based on measured values. Simultaneously, by fusing the mechanism prediction values ​​with data compensation terms, the physical constraints of the desulfurization reaction and gas-liquid mass transfer process are preserved, while operational data is used to compensate for model deviations, improving the accuracy and adaptability of SO2 concentration prediction under complex and variable operating conditions. Furthermore, by correcting the slurry pH setpoint based on outer-loop control commands and outputting slurry flow control commands and circulating pump frequency control commands, it helps to reduce excessive slurry supply and unnecessary circulating pump operation while ensuring SO2 emissions meet standards, thereby improving the closed-loop control stability and operational economy of the wet desulfurization system. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the SO2 concentration prediction and closed-loop control method of this application.

[0020] Figure 2 This is a schematic diagram of the SO2 concentration advance prediction and closed-loop control system of this application.

[0021] Figure 3 This is the overall flowchart of the method in this application.

[0022] Figure 4 A block diagram for constructing the hybrid model of this application is provided.

[0023] Figure 5 This is a diagram of a dual closed-loop control architecture consisting of an outer loop for SO2 concentration and an inner loop for pH.

[0024] Figure 6 This is a flowchart illustrating the dynamic weight allocation and economic optimization switching logic of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0026] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0027] In wet flue gas desulfurization (FGD) systems, SO2 concentration control is typically related to multiple factors, including flue gas load, inlet SO2 concentration, slurry pH, slurry density, circulating pump operating status, and limestone slurry supply status. Wet FGD systems inherently possess large inertia, pure time delay, strong nonlinearity, and strong multivariate coupling characteristics. Changes in outlet SO2 concentration often lag behind inlet flue gas disturbances, slurry reaction state changes, and actuator actions. In existing engineering applications, wet FGD SO2 concentration control often uses the measured SO2 concentration at the absorber outlet as feedback, and adjusts the slurry flow rate, circulating pump frequency, or related actuators via PID control. While this control method can maintain basic regulation under stable operating conditions, because the measured outlet SO2 value already includes the time delays caused by mass transfer, reaction, and detection links, rapid changes in inlet SO2 concentration, flue gas flow rate, unit load, or circulating pump operating status can cause control actions to lag behind actual desulfurization demands, leading to problems such as short-term SO2 exceedances at the outlet, excessive limestone slurry addition, and high circulating pump energy consumption.

[0028] On the other hand, existing SO2 concentration prediction methods can generally be divided into mechanistic model prediction and data-driven model prediction. Mechanistic models rely on desulfurization reaction kinetics and gas-liquid mass transfer theory, and have good physical interpretability. However, during long-term operation, wet desulfurization systems are affected by slurry aging, salt accumulation, uneven flow field, equipment characteristic drift, and unmodeled disturbances. Simple mechanistic models are difficult to continuously adapt to complex field conditions, and long-term prediction accuracy is prone to decline. Data-driven models can learn the nonlinear relationship between variables using historical operating data, but their prediction process lacks clear physical constraints. Under extreme or limited sample conditions such as variable load, circulation pump start-up and shutdown, and rapid fluctuations in inlet SO2 concentration, they are prone to insufficient generalization ability and low extrapolation reliability. Therefore, existing technologies still lack an engineering control method that can deeply integrate the interpretability of mechanistic models with the fitting ability of data models, and further combine it with SO2 concentration advance prediction, closed-loop control, and economic optimization.

[0029] Based on the above, in one embodiment of this application, the operating parameters of the wet desulfurization system are acquired in real time and preprocessed. A mechanism prediction value with physical constraints is generated using an SO2 concentration mechanism model. Then, a data-driven deviation compensation model is constructed based on the dynamic deviation between the mechanism prediction value and the measured outlet SO2 concentration. Finally, the mechanism prediction value and the data compensation term are merged into a mixed prediction value of the outlet SO2 concentration at a future prediction time. Further, an outer loop control for SO2 concentration is constructed using this mixed prediction value, and an inner loop control for pH is constructed using the slurry pH value. This allows the outer loop for SO2 concentration and the inner loop for pH to collaboratively generate control commands for the wet desulfurization system, thereby reducing feedback lag and improving prediction accuracy and control adaptability under complex operating conditions.

[0030] like Figure 1 The diagram shown is a flowchart illustrating a method for predicting and controlling SO2 concentration in advance, as proposed in this application. This application is used for predicting the SO2 concentration at the outlet of a wet desulfurization system and generating control commands. Wet desulfurization systems exhibit lag and coupling during operation. If only the measured value of the outlet SO2 concentration is relied upon for feedback adjustment, the control action often lags behind changes in flue gas load and absorption reaction state. This embodiment combines mechanism prediction, deviation compensation, and dual closed-loop control, enabling control commands to be generated before the actual outlet SO2 concentration deviates from the set target, thereby improving the problems of feedback lag and insufficient predictive adaptability under complex operating conditions. It may include: S101, real-time acquisition of the operating parameters of the wet desulfurization system, and preprocessing of the operating parameters; The operating parameters can come from on-site monitoring instruments, control systems, or data interfaces related to desulfurization operation in the wet desulfurization system. After acquiring the operating parameters, they are preprocessed to ensure that the data entering the subsequent prediction model meets the requirements for prediction calculation in terms of time sequence, data validity, and numerical expression. Preprocessing can reduce the impact of abnormal measurements, missing data, asynchronous sampling, and operating condition lags on the prediction results, providing a unified data foundation for subsequent mechanism prediction and data compensation.

[0031] S102, input the pre-processed operating parameters into the pre-established SO2 concentration mechanism model to obtain the mechanism prediction value of the outlet SO2 concentration; It should be noted that the SO2 concentration mechanism model is used to characterize the physical relationship between the operating state and the outlet SO2 concentration during wet desulfurization, and its output mechanism prediction value can serve as a physical benchmark for the trend of outlet SO2 concentration change. Since there may be incompletely modeled disturbances, equipment state drift, and changes in on-site operating conditions in the wet desulfurization system, relying solely on the mechanism model may not be sufficient to maintain adequate prediction accuracy over the long term. Therefore, this embodiment does not directly use the mechanism prediction value as the final prediction result, but further introduces dynamic deviation compensation.

[0032] S103, Based on the predicted value of the mechanism and the measured value of the outlet SO2 concentration, determine the dynamic deviation, and input the dynamic deviation into the pre-established data-driven deviation compensation model to obtain the data compensation item; It should be noted that the dynamic deviation reflects the discrepancy between the mechanistic model and the actual outlet SO2 concentration under current and historical operating conditions. After inputting the dynamic deviation into a pre-established data-driven deviation compensation model, the model learns and predicts the pattern of dynamic deviation change over time and outputs a data compensation term. This data compensation term is used to compensate for dynamic effects that the mechanistic model fails to accurately represent, ensuring that the prediction results simultaneously possess the physical consistency of the mechanistic model and the data model's ability to fit on-site disturbances.

[0033] S104, the mechanism prediction value is fused with the data compensation term to obtain the mixed prediction value of the outlet SO2 concentration at the future prediction time; It should be noted that after obtaining the mechanism prediction value and the data compensation term, the two are merged to obtain the mixed prediction value of the outlet SO2 concentration at the future prediction time. This mixed prediction value is not a simple reproduction of the current outlet SO2 concentration, but a forward prediction result formed for the future prediction time. By obtaining the future outlet SO2 concentration change trend before the control action is generated, a basis can be provided for the early adjustment of the wet desulfurization system, thereby reducing the impact of system pure time delay on the control effect.

[0034] S105, construct an outer loop control for SO2 concentration based on the predicted mixed value of SO2 concentration at the outlet, and construct an inner loop control for pH based on the pH value of the slurry, so that the outer loop control for SO2 concentration and the inner loop control for pH can work together to generate control commands for the wet desulfurization system.

[0035] It should be noted that during the control phase, an outer loop control for SO2 concentration is constructed based on the predicted mixed value of the outlet SO2 concentration, and an inner loop control for pH is constructed based on the slurry pH value. The outer loop control for SO2 concentration is used to determine the adjustment requirements for desulfurization capacity based on the future trend of outlet SO2 concentration changes, while the inner loop control for pH is used to match the slurry reactivity with the adjustment requirements of the outer loop. The outer and inner loops work together to generate control commands for the wet desulfurization system, ensuring that the control process focuses on both achieving the target outlet SO2 concentration and maintaining slurry pH stability. Thus, this embodiment improves upon the problems of lag in traditional feedback control, insufficient prediction accuracy of single models, and insufficient adaptability to complex operating conditions through a continuous process of "mechanism prediction—data compensation—mixing advance prediction—outer and inner loop collaborative control".

[0036] In one embodiment of this application, the SO2 concentration advance prediction and closed-loop control method is applicable to wet flue gas desulfurization systems in coal-fired power plants, especially suitable for large coal-fired units under complex operating conditions such as load changes, circulating pump start-up and shutdown, slurry state changes, slurry aging, salt accumulation, uneven flow field, or equipment characteristic drift. Wet flue gas desulfurization systems have characteristics of high inertia, pure time delay, strong nonlinearity, and strong coupling of multiple variables. Traditional PID control methods that use the measured value of outlet SO2 concentration as feedback are prone to only generating adjustment actions after actual emission changes, leading to short-term SO2 exceedances, excessive limestone slurry addition, or high circulating pump energy consumption. This embodiment provides a physical prediction benchmark using a mechanistic model, compensates for dynamic deviations in the mechanistic model with data-driven modeling, and constructs a collaborative control system for SO2 concentration outer loop and slurry pH inner loop based on mixed prediction results, achieving advance prediction, proactive adjustment, stable compliance, and economical operation.

[0037] like Figure 2 The diagram shown is a schematic representation of an SO2 concentration advance prediction and closed-loop control system according to this application, which may include: The data acquisition module is used to acquire the operating parameters of the wet desulfurization system in real time and to preprocess the operating parameters. The mechanism prediction module is used to input the pre-processed operating parameters into the pre-established SO2 concentration mechanism model to obtain the mechanism prediction value of the outlet SO2 concentration. The deviation compensation module is used to determine the dynamic deviation based on the predicted value of the mechanism and the measured value of the outlet SO2 concentration, and input the dynamic deviation into a pre-established data-driven deviation compensation model to obtain the data compensation item. The fusion prediction module is used to fuse the mechanism prediction value with the data compensation item to obtain the mixed prediction value of the outlet SO2 concentration at the future prediction time. The closed-loop control module is used to construct an outer loop control of SO2 concentration based on the predicted mixed value of SO2 concentration at the outlet, and to construct an inner loop control of pH based on the pH value of the slurry, so that the outer loop control of SO2 concentration and the inner loop control of pH work together to generate control commands for the wet desulfurization system.

[0038] like Figure 3 As shown, in this embodiment, the method is executed continuously according to the logic of real-time acquisition, preprocessing, mechanism modeling, deviation compensation, hybrid prediction, dual closed-loop control, command execution, and feedback optimization. First, key operating parameters of the unit and wet desulfurization system are acquired in real time through the DCS / OPC interface. These operating parameters include inlet flue gas parameters, slurry state parameters, control execution parameters, and controlled output parameters. Inlet flue gas parameters include inlet SO2 concentration, flue gas flow rate, boiler load, and flue gas temperature; slurry state parameters include slurry pH, slurry density, and slurry temperature; control execution parameters include slurry supply flow rate and circulating pump frequency; and controlled output parameters include the measured SO2 concentration at the absorber outlet. Optionally, combined with... Figure 4 The diagram shown illustrates the construction of the hybrid model. On the input side of the mechanistic model, limestone slurry concentration can also be collected, allowing the model to further reflect the impact of changes in absorbent supply capacity on the desulfurization reaction. When limestone slurry concentration is not a necessary input, it can be used as an extended operating parameter for model calibration or operational status diagnosis.

[0039] After acquiring the operating parameters, preprocessing is performed on them. Specifically, outlier removal can be performed on the collected data to reduce the impact of momentary instrument fluctuations or communication anomalies on the model input; bad points can be filtered to smooth measurements that clearly do not conform to the changing patterns of the operating conditions; missing values ​​can be interpolated to ensure that the continuous prediction process does not stop due to short-term data interruptions; time-series alignment can be performed on data from different sampling periods or different acquisition links to ensure that the inlet flue gas state, slurry state, execution state, and outlet SO2 concentration at the same calculation moment correspond in time; time-delay compensation can be performed on variables with transmission lag or process response lag to make the model input closer to the effective action time of the actual reaction process; and the model input variables can be normalized to ensure that data with different dimensions and numerical ranges can stably enter the data-driven deviation compensation model. Furthermore, validity judgment and redundancy verification of measuring points can be added. When key measuring points fail, exceed limits, or are inconsistent among multiple measuring points, invalid measuring points can be masked, replaced, or downweighted to ensure the quality of model input.

[0040] After data preprocessing, a SO2 concentration mechanism model was constructed. This model is based on desulfurization reaction kinetics and gas-liquid mass transfer theory, and is a mass transfer-reaction coupled model. During wet desulfurization absorption, SO2 in the flue gas enters the slurry phase and reacts with limestone slurry and oxygen. The core reaction can be represented as follows: CaCO3+SO2+1 / 2O2+2H2O→CaSO4·2H2O+CO2 Based on this reaction relationship and the gas-liquid mass transfer process, the mechanistic model uses inlet SO2 concentration, flue gas flow rate, boiler load, slurry pH, slurry density, overall mass transfer coefficient, and reaction rate constant as inputs or model parameters, and outputs a mechanistic prediction of outlet SO2 concentration. The mechanistic prediction can be expressed as: C_mech=f(SO2_in,Flow,Load,pH,Density,K_G,k) Wherein, C_mech is the mechanistic prediction value, SO2_in is the inlet SO2 concentration, Flow is the flue gas flow rate, Load is the boiler load, pH is the slurry pH value, Density is the slurry density, K_G is the overall mass transfer coefficient, and k is the reaction rate constant. K_G and k were determined through field operating tests, ensuring that the model parameters reflect the actual state of the absorber, slurry circulation, mass transfer conditions, and reaction capacity on site. This mechanistic model allows for baseline predictions of the outlet SO2 concentration while maintaining physical consistency.

[0041] like Figure 4 As shown, after the mechanism model module outputs the mechanism prediction value C_mech, a data compensation model module is further introduced. The data compensation model module receives the mechanism prediction value C_mech and the measured value of outlet SO2 C_actual, and calculates the dynamic deviation between them. The dynamic deviation satisfies: ΔC(t) = SO2_out_actual(t) C_mech(t) Wherein, ΔC(t) represents the dynamic deviation at time t, SO2_out_actual(t) is the measured SO2 concentration at the absorber outlet at time t, and C_mech(t) is the mechanism prediction value at time t. This dynamic deviation reflects the difference between the mechanism model and the actual operating conditions on site. The sources of this difference may include factors such as slurry aging, salt accumulation, uneven flow field, equipment characteristic drift, unmodeled disturbances, and sudden changes in operating conditions.

[0042] In one embodiment of this application, the data-driven deviation compensation model employs a time-series model combining VMD and LSTM. VMD performs variational mode decomposition on the dynamic deviation sequence, decomposing the original deviation sequence into multiple modal components with different frequency characteristics or change scales. The LSTM prediction sub-model learns the long-term time-series variation characteristics of each modal component individually or jointly, and generates corresponding prediction outputs. After weighted fusion, a data compensation term C_data is generated. Using this approach, LSTM can adapt to strongly nonlinear long-term time-series conditions, learning the slow-changing trends, periodic fluctuations, and nonlinear disturbances in the deviation sequence, thereby compensating for residual errors in the long-term predictions of the mechanistic model.

[0043] In another embodiment of this application, the data-driven deviation compensation model employs a time-series model combining VMD and LSSVM. VMD first performs mode decomposition on the dynamic deviation sequence, and then LSSVM learns and predicts the decomposed deviation features. This approach is suitable for rapidly changing operating conditions, such as rapid changes in unit load, start-up and shutdown of circulating pumps, sudden changes in flue gas flow, or rapid fluctuations in inlet SO2 concentration. LSSVM can quickly fit deviation changes within a short time window and output data compensation terms. Therefore, under rapidly changing operating conditions, LSSVM can be used to learn and predict dynamic deviations, while under strongly nonlinear long-term operating conditions, LSTM can be used to learn and predict dynamic deviations. The two models can be used individually or switched or combined based on the operating condition identification results.

[0044] To ensure the long-term effectiveness of the data compensation model, in some embodiments of this application, a model training period and a sliding window length can be set. The training period is used to determine the time interval for updating the parameters of the data-driven deviation compensation model, and the sliding window length is used to limit the range of historical data participating in the current deviation learning and prediction. When the sliding window is too short, the response capability to rapid changes in operating conditions can be enhanced; when the sliding window is too long, the ability to express long-term nonlinear trends can be enhanced. The training period and sliding window length can be configured according to the unit load change rate, the start-stop frequency of the circulating pump, the fluctuation characteristics of the outlet SO2 concentration, and the quality of the field data. Furthermore, abnormal deviation alarm and model freeze protection logic can be set. When the dynamic deviation is continuously abnormal, key measuring points fail, or the data distribution deviates significantly from the training sample range, the online update of the data compensation model is suspended or frozen to avoid abnormal data contaminating the model parameters.

[0045] After completing the mechanism prediction and data compensation, the model fusion module generates a mixed predicted value for the outlet SO2 concentration. The mixed predicted value satisfies: C_hybrid(t+τ)=C_mech(t+τ)+C_data(t+τ) Wherein, C_hybrid(t+τ) is the predicted mixed SO2 concentration at the outlet at the future prediction time, C_mech(t+τ) is the predicted mechanistic concentration at the future prediction time, C_data(t+τ) is the data compensation term at the future prediction time, and τ is the prediction lead time. This fusion relationship superimposes the physical baseline of the mechanistic model with the dynamic compensation of the data model, so that the prediction results do not deviate from the constraints of desulfurization reaction and gas-liquid mass transfer, while also incorporating unmodeled disturbances reflected in the field data.

[0046] In this embodiment, the prediction lead time τ is adaptively adjusted in real time based on the unit load, the number of circulating pumps in operation, and the absorber level. Changes in unit load affect the flue gas volume and inlet SO2 load, the number of circulating pumps in operation affects the slurry circulation intensity and gas-liquid contact conditions, and the absorber level affects the reaction residence time and liquid-gas contact state within the tower. By adjusting τ in combination with the above variables, the prediction lead time can be matched with the current lag characteristics of the wet desulfurization system. The value of τ ranges from 15s to 60s, used to achieve advance prediction of the outlet SO2 concentration 15s to 60s in the future, thereby allowing for advance control actions before the system outlet SO2 concentration actually exceeds the standard.

[0047] like Figure 5 As shown, after obtaining the predicted SO2 concentration C_hybrid, a dual closed-loop control system with an outer loop for SO2 concentration and an inner loop for slurry pH is constructed. This control system adopts an outer-loop-dominated, inner-loop-coordinated, dynamic weighted, and economically optimized architecture. The outer loop control of SO2 concentration uses the predicted SO2 concentration C_hybrid at the outlet as the controlled variable, compares C_hybrid with the SO2 emission setpoint C_set, and outputs a circulating pump frequency adjustment command based on the comparison result. When the predicted SO2 concentration at the outlet shows an increasing trend, the outer loop can improve the desulfurization capacity in advance, such as by increasing the circulating pump frequency or increasing the slurry circulation intensity; when the predicted SO2 concentration is lower than the set target and there is room for economic optimization, the outer loop can reduce unnecessary circulating pump energy consumption. By using the advanced predicted value as the basis for outer loop control, the control action no longer completely depends on the lagging measured outlet value, thereby reducing the control deviation caused by the pure time delay of the wet desulfurization system.

[0048] The pH inner loop control is used to maintain the slurry reactivity and receives control commands from the SO2 concentration outer loop control. Specifically, the pH inner loop control dynamically corrects the pH setpoint based on the outer loop control commands, then compares the corrected pH setpoint with the measured slurry pH value, and generates a limestone slurry supply flow command through the controller. Optionally, the pH inner loop control can use a PID controller to achieve stable pH adjustment of the slurry. The circulation pump frequency adjustment command output by the outer loop mainly affects slurry circulation and gas-liquid contact intensity, while the limestone slurry supply flow command output by the inner loop mainly affects slurry pH and absorption reaction capacity. Both work together in the wet desulfurization system to create a synergistic relationship between outlet SO2 concentration control, slurry reactivity maintenance, and operational economy.

[0049] like Figure 6 As shown, dynamic weight allocation can be performed based on SO2 prediction deviation when generating control commands. As an example, the SO2 prediction deviation can be determined according to the absolute deviation between the mixed prediction value and the SO2 emission setpoint, i.e.: ΔC=|C_hybrid C_set| When the SO2 prediction deviation is greater than 10 mg / m³ 3 When the SO2 concentration is within the target range, the system enters a compliance-priority mode, increasing the weight of SO2 concentration control and strengthening adjustment and correction to prioritize achieving the target SO2 concentration at the outlet. At this time, the control system can enhance the response of the circulating pump frequency adjustment and simultaneously correct the pH set point and slurry flow rate to improve desulfurization reaction capacity. When the SO2 prediction deviation is less than or equal to 10 mg / m³... 3 At this point, the system enters an economic optimization mode, reducing the weight of SO2 concentration control and minimizing unnecessary adjustments while ensuring emissions meet standards. In this state, multiple objectives can be optimized, such as SO2 compliance, pH stability, and gypsum quality standards, to reduce plant power consumption and limestone consumption.

[0050] In the economic optimization process, SO2 compliance constraints are used to limit the mixed predicted and measured values ​​of SO2 concentration at the outlet to not exceeding the emission control target; pH stability constraints are used to avoid frequent and large fluctuations in the slurry pH setpoint, preventing insufficient reactivity or excessive limestone addition; gypsum quality compliance constraints are used to prevent the quality of by-product gypsum from being affected by long-term high or low pH or unstable slurry supply. The control objects of multi-objective optimization include the circulation pump frequency and pH setpoint range, and the control output can include circulation pump frequency commands, pH setpoint correction commands, and slurry flow rate adjustment commands. Through this dynamic weight allocation and economic optimization switching logic, this embodiment can prioritize compliance when SO2 deviation is large, and take into account energy saving and consumption reduction when SO2 deviation is small.

[0051] During the instruction execution and online adaptive optimization phase, the control system outputs circulating pump frequency commands, pH setpoint correction commands, and slurry flow rate adjustment commands to the corresponding actuators. After the actuators operate, the measured SO2 concentration at the absorber outlet and the measured slurry pH value are fed back in real time to the hybrid model formed by the SO2 concentration mechanism model and the data-driven deviation compensation model. The hybrid model updates the dynamic deviation sequence based on the feedback data and updates the parameters of the data-driven deviation compensation model online, enabling the model to adapt to changes in slurry state, equipment performance, and operating condition distribution during long-term operation. Optionally, the system is configured with a periodic self-verification and parameter self-calibration process to periodically check and correct K_G, k, data compensation model parameters, deviation thresholds, or data normalization parameters to maintain long-term prediction and control accuracy.

[0052] In some embodiments of this application, safety protection logic can also be set. This safety protection logic includes at least one of the following: abnormal deviation alarm, model freeze protection, model failure determination, fail-safe logic, and a non-disruptive PID switchback mechanism. When the deviation between the measured outlet SO2 concentration and the mixed predicted value continuously exceeds a preset range, an abnormal deviation alarm is triggered; when a key measuring point fails, communication is interrupted, or the model input quality does not meet requirements, model freeze protection is triggered, suspending online updates of model parameters; when the model prediction result deviates from the actual output for a long period or the compensation model does not meet online operating conditions, a model failure determination is performed; when the prediction model or control command generation process is abnormal, fail-safe logic is executed to keep the wet desulfurization system in a safe operating state; when it is necessary to exit the mixed prediction closed-loop control, the original control mode is switched back through the non-disruptive PID switchback mechanism to avoid sudden changes in control output. This safety protection logic enables this application to be compatible with existing DCS systems and meets the requirements of reliability and continuous operation in engineering sites.

[0053] In one embodiment of this application, the above method can be executed by an SO2 concentration advance prediction and closed-loop control system. This system includes a data acquisition module, a mechanism prediction module, a deviation compensation module, a fusion prediction module, and a closed-loop control module. The data acquisition module acquires the operating parameters of the wet desulfurization system in real time through the DCS / OPC interface and preprocesses the operating parameters. The mechanism prediction module inputs the preprocessed operating parameters into a pre-established SO2 concentration mechanism model to obtain the mechanism prediction value of the outlet SO2 concentration. The deviation compensation module determines the dynamic deviation based on the mechanism prediction value and the measured value of the outlet SO2 concentration, and inputs the dynamic deviation into a pre-established data-driven deviation compensation model to obtain a data compensation term. The fusion prediction module fuses the mechanism prediction value and the data compensation term to obtain a mixed prediction value of the outlet SO2 concentration at a future prediction time. The closed-loop control module constructs an outer loop control for SO2 concentration based on the mixed prediction value of the outlet SO2 concentration and an inner loop control for pH based on the slurry pH value, enabling the outer loop control of SO2 concentration and the inner loop control of pH to collaboratively generate control commands for the wet desulfurization system.

[0054] Furthermore, the closed-loop control module includes an outer loop control unit for SO2 concentration and an inner loop control unit for pH. The outer loop control unit for SO2 concentration compares the predicted SO2 concentration at the outlet with the setpoint for SO2 emissions and outputs a circulating pump frequency adjustment command based on the comparison result. The inner loop control unit for pH receives the control command from the outer loop control unit for SO2 concentration and dynamically adjusts the pH setpoint accordingly to coordinate the limestone slurry supply flow rate with the adjustment requirements of the outer loop for desulfurization capacity. The system may also include an online optimization module and a safety protection module. The online optimization module feeds back the measured SO2 concentration at the outlet and the measured pH value of the slurry to the mixing model and updates the parameters of the data-driven deviation compensation model online. The safety protection module executes at least one of the following: abnormal deviation alarm, model freeze protection, model failure determination, fault-tolerant logic, and non-disruptive PID switching mechanism, to ensure that the predictive control process still has a safe and controllable operating boundary under abnormal conditions.

[0055] Through the above embodiments, this application provides a physical benchmark using a mechanistic model, compensates for dynamic deviations using a data-driven deviation compensation model formed by VMD and LSTM and / or VMD and LSSVM, and generates a mixed predicted value of the outlet SO2 concentration at future prediction times using C_hybrid(t+τ)=C_mech(t+τ)+C_data(t+τ). Control commands are then generated collaboratively using an outer loop for SO2 concentration and an inner loop for pH. Compared to feedback control that solely relies on measured outlet SO2 values, this application can obtain the trend of outlet SO2 concentration changes 15s to 60s in advance, reducing control lag. Compared to a single mechanistic model or a single data model, this application can simultaneously utilize physical consistency and data compensation capabilities to improve predictive adaptability under complex operating conditions. Through dynamic weight allocation and economic optimization, this application can reduce circulating pump energy consumption and limestone consumption while ensuring SO2 compliance, pH stability, and qualified gypsum quality.

[0056] In practical applications, this application can be deployed in wet desulfurization systems of large coal-fired power units. The system receives real-time data via a DCS / OPC interface, including inlet SO2 concentration, flue gas flow rate, boiler load, flue gas temperature, slurry pH, slurry density, slurry temperature, slurry supply flow rate, circulating pump frequency, and measured SO2 concentration at the absorber outlet. After outlier removal, bad point filtering, missing value interpolation, time sequence alignment, time delay compensation, normalization, measurement point validity assessment, and redundancy verification, the data is fed into the SO2 concentration mechanism model and the data-driven deviation compensation model.

[0057] As an example, when the unit load changes rapidly or the number of circulating pumps in operation changes, the flue gas flow rate, inlet SO2 load, and gas-liquid contact state of the wet desulfurization system will change synchronously, while the measured outlet SO2 concentration will show a lag. In this case, the mixing model first outputs C_mech from the mechanistic model, and then calculates the concentration based on ΔC(t) = SO2_out_actual(t). C_mech(t) forms a dynamic deviation sequence, and under rapid changing operating conditions, a time-series model combining VMD and LSSVM is invoked to predict C_data. Subsequently, the predicted outlet SO2 concentration at time τ is obtained according to C_hybrid(t+τ) = C_mech(t+τ) + C_data(t+τ), where τ is adaptively tuned based on unit load, the number of circulating pumps in operation, and the absorber level, with a value ranging from 15s to 60s. Using this advanced prediction result, the outer loop of SO2 concentration can output circulating pump frequency adjustment commands in advance, while the inner loop of pH synchronously corrects the pH setpoint and adjusts the limestone slurry flow rate, thereby enhancing desulfurization capacity before the actual increase in outlet SO2 concentration.

[0058] As another example, in situations where slurry aging, salt accumulation, or equipment characteristic drift leads to increased long-term prediction deviations in the mechanistic model, a time-series model combining VMD and LSTM can be used to learn dynamic deviations. VMD decomposes the deviation sequence into different modal components, and the LSTM prediction sub-model learns the long-term time-series variation characteristics of each modal component, which are then weighted and fused to obtain C_data. This data compensation term compensates for the mechanistic model's lack of dynamic modeling, allowing the mixed prediction value to be gradually and adaptively corrected according to the on-site operating conditions. The online optimization module continuously feeds back the measured values ​​of outlet SO2 concentration and slurry pH to the mixed model and updates the data-driven deviation compensation model parameters online to maintain long-term operational accuracy.

[0059] As an application example of dynamic weight allocation, when the SO2 prediction bias ΔC = |C_hybrid C_set|greater than 10mg / m 3 When SO2 levels are within acceptable limits, the control system enters a compliance-priority mode, increasing the weight of SO2 control and strengthening the coordinated correction of circulating pump frequency, pH setpoint, and slurry flow rate to prioritize SO2 compliance. When the SO2 prediction deviation ΔC is less than or equal to 10 mg / m³... 3 At this time, the control system enters the economic optimization mode, reduces the SO2 control weight, and performs multi-objective optimization of the circulating pump frequency and pH setting range under the constraints of SO2 compliance, pH stability and gypsum quality compliance, in order to reduce plant power consumption and limestone consumption.

[0060] As an application example of safe operation, when the dynamic deviation increases abnormally, the validity judgment of key measuring points fails, the redundancy verification is inconsistent, or the model output deviates from the measured outlet SO2 concentration for a long period of time, the system triggers an abnormal deviation alarm, model freeze protection, or model failure judgment. If the hybrid predictive closed-loop control does not meet the conditions for continued operation, the fault-safe logic is executed, and the original control mode is switched back through the non-disruptive PID switchback mechanism to avoid the sudden change of control command causing disturbance to the wet desulfurization system.

[0061] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for predicting and controlling SO2 concentration in advance, characterized in that, include: The operating parameters of the wet desulfurization system are acquired in real time, and the operating parameters are preprocessed. The pre-processed operating parameters are input into the pre-established SO2 concentration mechanism model to obtain the mechanism prediction value of the outlet SO2 concentration; Based on the predicted value of the mechanism and the measured value of the outlet SO2 concentration, the dynamic deviation is determined, and the dynamic deviation is input into a pre-established data-driven deviation compensation model to obtain the data compensation term. The mechanism prediction value is fused with the data compensation term to obtain the mixed prediction value of the outlet SO2 concentration at the future prediction time. An outer loop control for SO2 concentration is constructed based on the predicted mixed value of the outlet SO2 concentration, and an inner loop control for pH is constructed based on the pH value of the slurry, so that the outer loop control for SO2 concentration and the inner loop control for pH can work together to generate control commands for the wet desulfurization system.

2. The method according to claim 1, characterized in that, The operating parameters include inlet flue gas parameters, slurry state parameters, control execution parameters, and controlled output parameters; The inlet flue gas parameters include inlet SO2 concentration, flue gas flow rate, boiler load, and flue gas temperature; The slurry state parameters include slurry pH value, slurry density, and slurry temperature; The control execution parameters include slurry flow rate and circulation pump frequency; The controlled output parameters include the measured value of SO2 concentration at the absorber outlet.

3. The method according to claim 1, characterized in that, Preprocessing the operating parameters includes at least one of the following: outlier removal, bad point filtering, missing value interpolation, timing alignment, time delay compensation, normalization, measurement point validity judgment, and redundancy verification.

4. The method according to claim 1, characterized in that, The SO2 concentration mechanism model is a mass transfer-reaction coupling model established based on desulfurization reaction kinetics and gas-liquid mass transfer theory; The SO2 concentration mechanism model takes at least the inlet SO2 concentration, flue gas flow rate, boiler load, slurry pH value, slurry density, overall mass transfer coefficient and reaction rate constant as inputs or model parameters, and outputs the mechanism prediction value. The overall mass transfer coefficient and the reaction rate constant are determined through field operating condition tests.

5. The method according to claim 1, characterized in that, The dynamic deviation is the difference between the measured value of the outlet SO2 concentration and the mechanism prediction value at the corresponding time. The data-driven bias compensation model includes a time series model formed by combining VMD and LSTM, and / or a time series model formed by combining VMD and LSSVM.

6. The method according to claim 5, characterized in that, Under rapidly changing operating conditions, LSSVM is invoked to learn and predict the dynamic deviation, while under strongly nonlinear long-term operating conditions, LSTM is invoked to learn and predict the dynamic deviation.

7. The method according to claim 1, characterized in that, The predicted mixed SO2 concentration at the outlet satisfies: C_hybrid(t+τ)=C_mech(t+τ)+C_data(t+τ); Where C_hybrid(t+τ) is the predicted mixed SO2 concentration at the outlet at the future prediction time, C_mech(t+τ) is the predicted mechanistic concentration at the future prediction time, C_data(t+τ) is the data compensation term at the future prediction time, and τ is the prediction lead time; The prediction lead time τ is adaptively adjusted in real time based on the unit load, the number of circulating pumps in operation, and the liquid level in the absorption tower, and τ is 15s to 60s.

8. The method according to claim 1, characterized in that, The outer loop control of SO2 concentration includes: comparing the predicted value of SO2 concentration at the outlet with the set value of SO2 emission, and outputting a circulating pump frequency adjustment command based on the comparison result; The pH inner loop control includes: receiving the control command of the SO2 concentration outer loop control, and dynamically correcting the pH setpoint according to the control command; When generating control commands for the wet desulfurization system, dynamic weight allocation is performed based on SO2 prediction deviation, and multi-objective optimization is carried out on the circulation pump frequency and pH setting range, with SO2 compliance, pH stability and gypsum quality as constraints. Among them, when the SO2 prediction deviation is greater than 10 mg / m³ 3 When SO2 concentration control weight is increased; when SO2 prediction deviation is less than or equal to 10 mg / m³ 3 At that time, the SO2 concentration control weight should be reduced.

9. A SO2 concentration advance prediction and closed-loop control system, characterized in that, include: The data acquisition module is used to acquire the operating parameters of the wet desulfurization system in real time and to preprocess the operating parameters. The mechanism prediction module is used to input the pre-processed operating parameters into the pre-established SO2 concentration mechanism model to obtain the mechanism prediction value of the outlet SO2 concentration. The deviation compensation module is used to determine the dynamic deviation based on the predicted value of the mechanism and the measured value of the outlet SO2 concentration, and input the dynamic deviation into a pre-established data-driven deviation compensation model to obtain the data compensation item. The fusion prediction module is used to fuse the mechanism prediction value with the data compensation item to obtain the mixed prediction value of the outlet SO2 concentration at the future prediction time. The closed-loop control module is used to construct an outer loop control of SO2 concentration based on the predicted mixed value of SO2 concentration at the outlet, and to construct an inner loop control of pH based on the pH value of the slurry, so that the outer loop control of SO2 concentration and the inner loop control of pH work together to generate control commands for the wet desulfurization system.

10. The system according to claim 9, characterized in that, The data acquisition module acquires the operating parameters in real time through the DCS / OPC interface; The closed-loop control module includes an outer loop control unit for SO2 concentration and an inner loop control unit for pH. The outer loop control unit for SO2 concentration is used to compare the predicted value of SO2 concentration at the outlet with the set value of SO2 emission, and output a circulating pump frequency adjustment command based on the comparison result. The inner loop control unit for pH is used to receive the control command from the outer loop control unit for SO2 concentration, and dynamically correct the pH set point based on the control command. The system also includes an online optimization module and a safety protection module. The online optimization module is used to feed back the measured values ​​of outlet SO2 concentration and slurry pH to the hybrid model formed by the SO2 concentration mechanism model and the data-driven deviation compensation model, and update the parameters of the data-driven deviation compensation model online. The safety protection module is used to execute at least one of the following: abnormal deviation alarm, model freeze protection, model failure judgment, fault-safe logic, and non-intrusive PID switchback mechanism.

Images