An SCR denitration control method, system, medium and program product based on load prediction feedforward and estimation compensation feedback

Abstract

The present application relates to the technical field of thermal power generator set control, in particular to a SCR denitration control method and system based on load prediction feedforward and estimation compensation feedback, medium and program product, the method comprises: predicting flue gas temperature and flow through boiler load signal, establishing a theoretical ammonia demand calculation model, calculating the theoretical ammonia feedforward demand; based on the ammonia injection valve opening, ammonia injection amount and outlet NOx concentration historical data identification to establish the dynamic characteristic model of SCR denitration system; build an estimation compensation control framework, use the internal model output and actual output deviation to form estimation error and calculate ammonia injection correction amount; according to the ammonia feedforward demand and feedback correction amount to generate ammonia injection valve control instruction, lead control ammonia injection valve opening and closing, quickly adjust the outlet NOx concentration. The present application can improve the large lag problem existing in denitration control, predict the change of valve opening, achieve advanced regulation and realize the optimization control of SCR denitration system.

Description

Technical Field

[0001] This invention belongs to the field of thermal power generator control technology, and relates to the regulation of ammonia injection in SCR systems, control of outlet NOx concentration, and dynamic compensation control with time delay. More specifically, it relates to an SCR denitrification control method, system, medium, and program product based on load prediction feedforward and estimated compensation feedback. Background Technology

[0002] Selective catalytic reduction (SCR) is a commonly used flue gas purification method for coal-fired power units. It typically includes an ammonia source and metering injection device, an ammonia injection grid and mixing section, a reactor and catalyst bed, flue gas temperature / flow measurement, an outlet NOx analyzer, and a distributed control system (DCS). During operation, the ammonia injection rate is the primary controlled variable, causing the reducing agent to react with NOx on the catalyst surface, thereby meeting emission constraints.

[0003] The SCR process exhibits significant hysteresis and inertia in engineering, accompanied by nonlinear, strongly coupled, and time-varying characteristics: the applicable flue gas temperature range, space velocity, inlet NOx concentration, oxygen content, catalyst aging / poisoning, and dust load all alter reaction kinetics; uneven mixing and distribution cause spatial differences in stoichiometry, affecting conversion efficiency; simultaneously, the sampling transmission link and flue gas stagnation introduce pure hysteresis, resulting in a long delay between ammonia injection adjustment and outlet NOx response. These uncertainties are further amplified when the unit load increases or decreases, combustion fluctuates, and coal quality changes.

[0004] Current engineering control systems mostly employ conventional PID feedback regulation with simple compensations such as inlet NOx and flue gas flow rate. While this method is easy to implement, it often requires conservative tuning for objects with large time delays to avoid oscillations, leading to slow response. Under rapid load changes or strong disturbances, accumulated errors can easily cause short-term NOx exceedances, or excessive ammonia injection can lead to increased ammonia escape, increased risk of scaling and corrosion in the air preheater, and increased reducing agent consumption. Meanwhile, ammonia injection valves / metering devices have saturation, dead zone, and rate-of-change limitations; measurement noise and operating condition switching also weaken control quality.

[0005] To improve dynamic performance, the industry has proposed approaches such as disturbance compensation, cascade / hierarchical control, and model-based control. However, these approaches still face challenges in SCR applications: measurement lag and coupling of disturbance quantities lead to mismatch in compensation; model parameters drift with load range and catalyst state, making it difficult to apply fixed models in the long term; and the coexistence of multi-objective constraints (NOx, ammonia slip) and actuator limitations makes it difficult to balance speed and robustness. For example, Chinese patent CN120733553A discloses a data-driven ammonia injection prediction control method for SCR denitrification systems in coal-fired power plants. This method uses a BP neural network and a Long Short-Term Memory (LSTM) network to predict input parameters as feedforward. However, this type of method is highly dependent on high-quality historical sample data, and its model generalization ability is limited under non-steady-state processes with large operating conditions, making it difficult to guarantee prediction accuracy. CN120255343A discloses an unbiased model predictive control method for SCR denitrification systems based on dynamic variable gain Kalman filtering. While this method improves disturbance estimation speed by dynamically adjusting the gain, it primarily focuses on passive observation and compensation for existing disturbances, failing to establish an advanced prediction mechanism from the source of load changes. This results in significant phase mismatch between feedforward compensation and feedback regulation on the time scale when dealing with objects with extremely long time delays. These technical solutions still have room for improvement in handling the deep coordination between advanced trend prediction and large time delay prediction compensation.

[0006] In summary, existing SCR denitrification control systems still struggle to achieve rapid and stable control of outlet NOx under conditions of large lag, time-varying coupling, and constraints. Therefore, improving the dynamic control quality and robustness of the SCR denitrification process under varying load disturbances is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0007] (a) Purpose of the invention In view of the above-mentioned shortcomings and deficiencies of the prior art, this invention provides an SCR denitrification control method, system, medium, and program product based on load prediction feedforward and predictive compensation feedback. It collects relevant operating signals of the unit load and establishes predictive models for key operating parameters such as flue gas temperature and flow rate. Combined with inlet NOx and other measurement information, it constructs a basis for calculating ammonia injection demand. Simultaneously, based on the dynamic characteristics identification results of the SCR denitrification process, it establishes a controlled object model containing pure time delay. On this basis, it constructs a predictive compensation feedback control framework to output ammonia injection correction. The feedforward ammonia injection quantity and the feedback correction quantity are fused to generate ammonia injection valve control commands, achieving rapid tracking and disturbance suppression of outlet NOx concentration. This addresses the problem that existing coal-fired unit SCR denitrification systems have large delays and inertia in ammonia injection control, making it difficult to achieve NOx control. x The issue of precise emission control.

[0008] (II) Technical Solution To achieve the objective of this invention and solve its technical problems, the present invention adopts the following technical solution: The first objective of this invention is to provide an SCR denitrification control method based on load prediction feedforward and prediction compensation feedback, comprising at least the following steps: SS1. Construct a load-flue gas parameter prediction model: Obtain historical operating data of the unit, including at least boiler load and corresponding fuel supply and air supply, as well as SCR inlet flue gas temperature and flue gas flow rate, and establish a load-flue gas parameter prediction model based on historical operating data; SS2. Flue gas parameters and denitrification efficiency prediction: Real-time acquisition of the current unit load command and its rate of change, as input variables for the load-flue gas parameter prediction model, to predict the SCR inlet flue gas temperature and flue gas flow rate and their changing trends within a preset time window, and to determine the denitrification efficiency η corresponding to the predicted flue gas temperature; SS3. Construct a theoretical ammonia demand calculation model: Based on the predicted flue gas flow rate and the denitrification efficiency η corresponding to the predicted flue gas temperature, combined with the real-time collected SCR inlet NOx concentration and ammonia nitrogen molar ratio NSR, establish a theoretical ammonia demand calculation model, calculate the theoretical ammonia demand to meet the future instantaneous operating conditions, and form the ammonia injection feedforward control quantity. SS4. Establish SCR dynamic characteristic model: Obtain historical operating data of SCR denitrification system, including at least ammonia injection valve opening, ammonia injection rate, and outlet NOx concentration; based on the historical operating data of SCR denitrification system, identify parameters, establish a first-order ammonia flow dynamic characteristic sub-model with ammonia injection valve opening as input and ammonia injection rate as output, and a second-order outlet NOx concentration dynamic characteristic sub-model with time delay with ammonia injection rate as input and outlet NOx concentration as output, and connect the two sub-models in series to form a third-order SCR dynamic characteristic model G(s) with time delay; SS5. Construct a predictive compensation feedback control framework: Design a predictive compensation feedback control framework based on G(s), including at least a low-pass filter F1(s), a phase compensation component F2(s), and a tracking controller G. c1 (s), Correction controller G c2 (s), Estimated internal model G m The (s) and its time delay compensation stage take the outlet NOx concentration setpoint r as input, construct the prediction error by the deviation between the internal model output and the actual output, and perform phase compensation on the prediction error through F2(s) to suppress the dynamic mismatch caused by time delay. The phase-compensated prediction error is then input into G. c2 (s) Generates correction control quantity, and finally outputs ammonia injection feedback correction quantity used to correct ammonia injection feedforward control quantity; SS6. Ammonia injection valve advance control: Based on the ammonia injection feedforward control quantity and the ammonia injection feedback correction quantity, an ammonia injection valve control command is generated to adjust the outlet NOx concentration.

[0009] The second objective of this invention is to provide an SCR denitrification control system, employing the aforementioned SCR denitrification control method based on load prediction feedforward and prediction compensation feedback, which includes at least a load-flue gas parameter prediction module, a theoretical ammonia demand calculation module, an SCR dynamic characteristic modeling module, a prediction compensation feedback control module, and an ammonia injection valve advance control module, wherein: The load-flue gas parameter prediction module is used to acquire historical operating data of the unit, including at least boiler load and corresponding fuel and air supply, as well as SCR inlet flue gas temperature and flow rate, and to establish a load-flue gas parameter prediction model based on the historical operating data of the unit. The flue gas parameter and denitrification efficiency prediction module is used to collect the current unit load command and its rate of change in real time as model input, predict the SCR inlet flue gas temperature and flow rate and their changing trend within a preset time window, and determine the denitrification efficiency η corresponding to the predicted flue gas temperature. The theoretical ammonia demand calculation module is used to establish a theoretical ammonia demand calculation model based on the predicted flue gas flow rate and denitrification efficiency η, combined with the real-time collected SCR inlet NOx concentration and ammonia nitrogen molar ratio NSR, to calculate the theoretical ammonia demand that meets the future instantaneous operating conditions and form the ammonia injection feedforward control quantity. The SCR dynamic characteristic modeling module is used to acquire historical operating data of the SCR denitrification system, including at least the opening degree of the ammonia injection valve, the amount of ammonia injected, and the outlet NOx concentration. It establishes a first-order ammonia flow dynamic characteristic sub-model G2(s) with the opening degree of the ammonia injection valve as input and the amount of ammonia injected as output, and a second-order outlet NOx concentration dynamic characteristic sub-model G1(s) with the amount of ammonia injected as input and the outlet NOx concentration as output. G1(s) and G2(s) are connected in series to form a third-order SCR dynamic characteristic model G(s) with time delay. The prediction compensation feedback control module includes a low-pass filter F1(s), a phase compensation circuit F2(s), and a tracking controller G. c1 (s), Correction controller G c2 (s), Estimated internal model G m The (s) and its time delay compensation stage take the outlet NOx concentration setpoint r as input, construct the prediction error by the deviation between the internal model output and the actual output, and perform phase compensation on the prediction error through F2(s) to suppress the dynamic mismatch caused by time delay. The phase-compensated prediction error is then input into G. c2 (s) Generate correction control quantity and output ammonia injection feedback correction quantity; The ammonia injection valve advance control module is used to generate ammonia injection valve control commands based on the ammonia injection feedforward control quantity and the ammonia injection feedback correction quantity, so as to adjust the ammonia injection quantity and realize the outlet NOx concentration control.

[0010] The second objective of this invention is to provide a computer program product, including computer instructions, for executing the steps of the above-described SCR denitrification control method based on load prediction feedforward and prediction compensation feedback of this invention.

[0011] The third objective of this invention is to provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described SCR denitrification control method based on load prediction feedforward and prediction compensation feedback.

[0012] (III) Technical Effects Compared with the prior art, the SCR denitrification control method, system, medium, and program products based on load prediction feedforward and prediction compensation feedback of the present invention have the following beneficial and significant technical effects: (1) The present invention calculates the theoretical ammonia injection feedforward demand based on the flue gas flow and temperature obtained by boiler load prediction, and calculates the ammonia injection feedback correction amount through the Smith predictive feedback controller with phase compensation, which can advance the control of the opening and closing of the ammonia injection valve and realize the rapid adjustment of the outlet NOx concentration.

[0013] (2) The present invention innovatively constructs a Smith predictive controller for the denitrification system. This controller has better setpoint tracking capability and anti-disturbance capability. The load prediction coupling phase compensation link can improve the problem of large lag in denitrification control, compensate for pure delay time, improve the stability and response speed of the controlled object, and achieve advanced adjustment.

[0014] (3) This invention uses feedforward and prediction compensation feedback in combination for multi-constraint ammonia injection control, which can suppress the risk of ammonia escape and air preheater scaling caused by excessive ammonia injection while meeting the NOx control accuracy, and reduce the high-frequency invalid valve action; combined with the online update mechanism of model parameters, it can adapt to catalyst activity decay and operating condition drift, and improve long-term operation economy and reliability. Attached Figure Description

[0015] Figure 1 This is a flowchart of the control method of the present invention; Figure 2 This is an architecture diagram of the control method of the present invention; Figure 3 This is a structural block diagram of the predictive compensation feedback control framework in this invention; Figure 4 This is a flowchart of parameter tuning for each stage of the predictive compensation feedback control framework in this invention. Detailed Implementation

[0016] This invention aims to provide a method, system, medium, and program product for SCR denitrification control based on load prediction feedforward and prediction compensation feedback. To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions in the embodiments of this invention will be described in more detail below with reference to the accompanying drawings. The described embodiments are some, but not all, embodiments of this invention, and are exemplary, intended to explain the invention, and should not be construed as limiting the invention. All other embodiments obtained based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0017] Example 1: Control Method See Figure 1 , Figure 2 The SCR denitrification control method based on load prediction feedforward and prediction compensation feedback provided in this embodiment of the invention is used to adjust the ammonia injection rate in the SCR denitrification system of a coal-fired power generating unit to control the outlet NOx concentration. This control method, when implemented, includes at least the following steps: SS1. Construct a load-flue gas parameter prediction model: Historical operating data of the unit is acquired, including at least boiler load and corresponding fuel and air supply, as well as SCR inlet flue gas temperature and flow rate. A load-flue gas parameter prediction model is then established based on this historical operating data. Before modeling, the historical operating data undergoes consistency verification and cleaning, including: time alignment of boiler load, fuel supply, air supply, SCR inlet flue gas temperature, and flow rate according to the sampling period; outlier removal and missing segments completion; smoothing and filtering of the acquired signals to suppress high-frequency noise; and normalization or standardization of each variable to form a training sample set, which is then divided into training and validation sets to evaluate generalization error.

[0018] In this embodiment of the invention, the load-flue gas parameter prediction model is a multi-input multi-output prediction model. It uses boiler load, fuel supply, and air supply as input variables, and SCR inlet flue gas temperature and flow rate as output variables. At least one load-related dynamic feature is introduced as an auxiliary input, which can be the load change rate or the fuel or air supply change rate. The prediction model employs a time-series prediction architecture based on a Long Short-Term Memory (LSTM) network, including: using boiler load, corresponding fuel and air supply, and the dynamic feature as input feature vectors of the LSTM network, and using the SCR inlet flue gas temperature and flow rate as output target variables; extracting the time-series evolution features of flue gas parameters through hidden layers, training the LSTM network weight parameters through backpropagation, and using a Dropout layer to prevent overfitting, thus converging the prediction error to a preset threshold range; after training, the load-flue gas parameter model predicts the dynamic change trend of the SCR inlet flue gas temperature and flow rate within the next 5-15 minutes based on current and historical load data.

[0019] In a preferred embodiment, the load-flue gas parameter prediction model employs a rolling update mechanism: the latest sampled data is continuously added during unit operation, and the model parameters are calibrated online. This allows the prediction model to adaptively adjust to changes in fuel quality, air and coal distribution, and operating mode, thereby reducing the impact of long-term drift on the prediction accuracy of flue gas temperature and flow rate. Furthermore, the rolling update can be triggered by model residuals or prediction confidence: when the error within the validation window exceeds a threshold or the residuals exhibit a systematic shift, incremental training is initiated and key layer parameters are frozen to balance stability and adaptability.

[0020] It should be noted that the load-flue gas parameter prediction model in this step, by introducing dynamic features such as the load change rate, can capture the transient coupling characteristics of the combustion process under varying load conditions, overcoming the prediction lag defect of the traditional steady-state mapping model during load disturbances; the memory gate mechanism of the LSTM network enables the model to retain long-term evolution information of the combustion system, and can still maintain prediction accuracy when coal quality fluctuates or air distribution strategy changes; the rolling update mechanism continuously absorbs operating data to correct model weights, realizing the adaptive tracking of the prediction model to catalyst activity decay, seasonal fuel quality changes and operation mode optimization, ensuring the accuracy and robustness of the feedforward control quantity throughout the entire life cycle.

[0021] SS2. Prediction of flue gas parameters and denitrification efficiency: The system collects real-time data on the current unit load command and its rate of change, using this data as input variables for a load-flue gas parameter prediction model. This model predicts the SCR inlet flue gas temperature and flow rate and their trends within a preset time window, and determines the denitrification efficiency η corresponding to the predicted flue gas temperature. The length of the preset time window is adaptively set based on the time delay parameter and dominant time constant of the SCR dynamic characteristic model G(s), ensuring that the preset time window covers the response range where changes in SCR inlet flue gas parameters significantly affect the outlet NOx concentration. The prediction results are updated in a rolling manner within each control cycle, providing a real-time updated prediction basis for the ammonia injection feedforward control quantity as the operating conditions change. Within the prediction time window, a sampling strategy with equal or variable time intervals is employed to match the distribution of prediction points with the SCR dynamic response characteristics.

[0022] In this embodiment of the invention, the method for determining the denitrification efficiency η is as follows: A mapping relationship database between flue gas temperature T and denitrification efficiency η is established in advance through SCR reactor catalyst activity test experiments. The mapping relationship is represented as a piecewise function. When T < 300℃, η increases exponentially with increasing temperature. When 300℃ ≤ T ≤ 420℃, η remains in the high-efficiency range of 0.80~0.95. When T > 420℃, η decreases due to the intensified ammonia oxidation reaction. After real-time prediction of flue gas temperature, the instantaneous denitrification efficiency η value under the corresponding operating condition is obtained by querying the mapping relationship database and performing linear interpolation calculation. When the predicted flue gas temperature exceeds the calibrated temperature range, η is subjected to boundary truncation or extrapolation constraint processing. When extrapolating, a conservative coefficient is introduced to reduce the η value of the extrapolation segment, so as to avoid unreasonable theoretical ammonia demand calculation results when the temperature deviates from the normal reaction window at low or high temperatures.

[0023] SS3. Construct a theoretical ammonia demand calculation model: Based on the predicted flue gas flow rate and the corresponding denitrification efficiency η corresponding to the predicted flue gas temperature, combined with the real-time collected SCR inlet NOx concentration and ammonia nitrogen molar ratio (NSR), a theoretical ammonia demand calculation model is established to calculate the theoretical ammonia demand that meets future instantaneous operating conditions, thus forming the ammonia injection feedforward control quantity.

[0024] The theoretical ammonia demand calculation model is based on flue gas flow rate Q and inlet NOx concentration C. NOx The denitrification efficiency η and the ammonia-to-nitrogen molar ratio NSR are used to construct a material balance relationship, and the molar mass ratio M of ammonia to NOx is introduced for unit conversion, so that the ammonia injection feedforward control quantity U is... ff Satisfy U ff =Q×C NOx ×η×M / NSR, and in the calculation process, the flue gas flow rate Q is converted from volume basis to molar flow rate or mass flow rate, NOx is converted to a uniform basis oxygen content or uniform dry basis conditions, and C is adjusted accordingly. NOx Moving average filtering is used to suppress measurement noise.

[0025] As a preferred option, the ammonia-nitrogen molar ratio (NSR) is adaptively adjusted based on the target outlet NOx concentration and predicted operating conditions. Ammonia slip or ammonia-nitrogen ratio constraints are introduced to form a constrained theoretical ammonia demand calculation: Under the premise of meeting the outlet NOx concentration control target, an upper limit is set for the NSR, and the theoretical ammonia demand is corrected by combining it with the reaction window corresponding to the predicted flue gas temperature, so that the ammonia injection feedforward control quantity U... ff The safety boundary is met under both high temperature and high activity or low temperature and low activity conditions, and the constraint information is transmitted to the correction channel of step SS5 when the boundary is triggered, so as to avoid feedback compensation from further increasing the amount of ammonia injected.

[0026] SS4. Establish the SCR dynamic characteristic model: Obtain historical operating data of the SCR denitrification system, including at least the opening degree of the ammonia injection valve, the ammonia injection rate, and the outlet NOx concentration. Based on the historical operating data of the SCR denitrification system, identify parameters and establish a first-order ammonia flow dynamic characteristic sub-model with the ammonia injection valve opening degree as input and the ammonia injection rate as output, and a second-order outlet NOx concentration dynamic characteristic sub-model with time delay with the ammonia injection rate as input and the outlet NOx concentration as output. Connect the two sub-models in series to form a third-order SCR dynamic characteristic model G(s) with time delay.

[0027] In this embodiment of the invention, parameter identification based on historical SCR operating data is used to construct a sub-model G1(s) of dynamic characteristics of outlet NOx concentration and a sub-model G2(s) of dynamic characteristics of ammonia flow rate, including the following sub-steps: S411. Acquisition of historical operating data: Conduct identification tests on the SCR denitrification system or collect historical operating data of the SCR denitrification system using available identification data windows formed during normal unit operation, including at least the opening degree of ammonia injection valve, ammonia injection rate, and outlet NOx concentration, and perform preprocessing on the data; S412. Identification of the Dynamic Characteristics Sub-model for Outlet NOx: Using ammonia injection rate as input and outlet NOx concentration as output, parameter identification is performed to establish a dynamic characteristics sub-model for outlet NOx concentration. Its transfer function is expressed using a second-order inertial element combined with a pure time delay structure, satisfying... Where K1 is the gain parameter, T1 and T2 are time constants, τ is the object pure time delay parameter of G(s), used to characterize the inertia and pure time delay characteristics of the response from the change in ammonia injection rate to the outlet NOx, and s is the Laplace operator. Indicates the delay stage; S413. Identification of the dynamic characteristic sub-model of ammonia flow rate: Using the opening degree of the ammonia injection valve as the input and the ammonia injection rate as the output, the transfer function of the dynamic characteristic sub-model of ammonia flow rate is established. Where K2 is the gain parameter, T3 is the time constant, and s is the Laplace operator; S414. Construction of SCR Dynamic Characteristic Model: The transfer functions G1(s) and G2(s) of the two sub-models are connected in series to form the SCR dynamic characteristic model. G0(s) is an SCR object used to characterize the overall dynamic transmission process from the change in the opening degree of the ammonia injection valve, through the amount of ammonia injected, to the change in the NOx concentration at the outlet, and serves as the object model for the subsequent prediction, compensation, and feedback control framework.

[0028] In a preferred embodiment, parameter identification employs the least squares method, recursive least squares method, or maximum likelihood method, combined with a forgetting factor mechanism to achieve online updates. Before identification, the input and output data undergo time synchronization, noise filtering, and outlier removal. A persistent excitation constraint is applied to the identified input and output data, including: applying a step excitation of preset amplitude and duration to the ammonia injection valve opening or ammonia injection quantity, provided that the valve opening does not trigger saturation and meets the rate of change limit, so that the ammonia injection valve opening or ammonia injection quantity has sufficient dynamic change amplitude and frequency band coverage within the identification time window; and performing correlation analysis and pulse / step excitation on historical data of ammonia injection quantity and outlet NOx concentration. The step response analysis determines the gain parameter K1, time constants T1 and T2, and time delay constant τ of G1(s) based on the dynamic response of NOx at the outlet to the ammonia injection rate. The time-series correlation and step response analysis are performed on the historical data of ammonia injection valve opening and ammonia injection rate. The gain parameter K2 and time constant T3 of G2(s) are determined based on the response curve of ammonia injection rate to valve opening. The forgetting factor λ is limited to (0,1) and is adaptively adjusted based on the model residual and the rate of change of operating conditions. This allows the parameters of G1(s) and G2(s) to be updated recursively during operation to adapt to the dynamic characteristic drift caused by catalyst activity decay, flue gas composition fluctuations, and load changes.

[0029] It should be noted that the tandem sub-model identification method used in this step decomposes the complex dynamics of the SCR system into an ammonia supply stage and a denitrification reaction stage, improving the physical interpretability of the model and the accuracy of parameter identification. The second-order inertia plus pure time delay structure can accurately describe the multi-scale dynamic processes of ammonia diffusion, catalytic reaction and flue gas transport in the SCR reactor, providing a high-fidelity object model for predictive compensation control. The persistent excitation constraint ensures that the identification data contains sufficient frequency domain information, avoiding parameter uncertainty caused by underexcitation. The adaptive adjustment mechanism of the forgetting factor enables the model to track characteristic drift caused by catalyst activity decay, coal type changes and operation mode adjustments, ensuring the robustness of the control system throughout its entire life cycle. The parameter physical constraints and rationality verification prevent parameter divergence or distortion during online updates, ensuring the long-term stability and reliability of the model.

[0030] SS5. Construct a predictive compensation feedback control framework: Based on G(s), a predictive compensation feedback control framework is designed, which includes at least a low-pass filter F1(s), a phase compensation component F2(s), and a tracking controller G. c1 (s), Correction controller G c2 (s), Estimated internal model G m The (s) and its time delay compensation stage take the outlet NOx concentration setpoint r as input, construct the prediction error by the deviation between the internal model output and the actual output, and perform phase compensation on the prediction error through F2(s) to suppress the dynamic mismatch caused by time delay. The phase-compensated prediction error is then input into G. c2 (s) Generates the correction control quantity, and finally outputs the ammonia injection feedback correction quantity used to correct the ammonia injection feedforward control quantity.

[0031] In this embodiment of the invention, the predictive compensation feedback control framework adopts an improved Smith predictor architecture, such as... Figure 3 As shown, its topology is as follows: taking the outlet NOx concentration setpoint r as input, r is processed by the low-pass filter F1(s) to obtain a smooth reference signal; based on the tracking controller G... c1 The output of (s) constructs a dual-channel prediction and correction structure, where: G c1 (s) One output of u1 is connected to the correction controller G at the object-driven synthesis node. c2 The output of (s) is algebraically synthesized to obtain the object driving signal U(s), and U(s) is combined with the inlet NOx perturbation D. NOx (s) is superimposed at the input of the object and used as the input of the SCR object G0(s), which then passes through G0(s) and the object pure time delay stage. After processing, the outlet NOx concentration output y(t) is obtained; G c1 (s) Output u1's other input to predict the internal model G m (s) Obtain the estimated intermediate quantity P m (s), and P m (s) After the internal model pure time delay stage Obtain the internal model output y m (t), where τ m To estimate the time delay constant of the internal model; compare y(t) with y m (t) The difference at the comparison node yields the estimated error e. m (t)=y(t)-y m (t); the estimated error e m (t) is processed by the phase compensation stage F2(s) and then input to the correction controller G. c2 (s), outputting the feedback correction amount u2 and participating in the synthesis of the object driving signal U(s); simultaneously, the prediction error e m (t) and the estimated intermediate value P m(s) The main bias signal is superimposed at the main bias construction node and compared with the smoothed reference signal to form the main bias signal, which serves as the tracking controller G. c1 The input of (s) forms a prediction feedback closed loop with phase compensation; the prediction internal model G m (s) adopts a dynamic model with the same order structure as G(s) but whose parameters can be adjusted online. The time delay constant τ of its time delay compensation element is... m Take G(s) as 0.95 to 1.05 times the time delay constant τ.

[0032] In this embodiment of the invention, the low-pass filter F1(s) is set on the input side of the outlet NOx concentration setpoint r or on the difference side between the setpoint and the feedback value, to limit the high-frequency control components caused by sudden changes in the setpoint. F1(s) is a first-order inertial low-pass filter, and its transfer function satisfies... , where τ f F1(s) is the filtering time constant, s is the Laplace operator, and the cutoff frequency of F1(s) is set according to the sampling period of the outlet NOx concentration and the noise spectrum characteristics of the measurement signal. This ensures that the filtered signal retains the dominant dynamics of the SCR denitrification system while suppressing high-frequency fluctuations caused by measurement noise and interference, and reducing the ineffective operation of the ammonia injection valve.

[0033] The phase compensation stage F2(s) is set on the prediction error channel formed by the internal model output and the actual output, so that the prediction error is phase compensated before being input to the correction controller G. c2 (s) to improve the speed of prediction compensation feedback in correcting the deviation of outlet NOx concentration and reduce overshoot during load changes; F2(s) adopts a second-order phase compensation network structure, based on G(s) and the prediction internal model G m The phase difference between (s) is designed to compensate for the phase lag caused by time delay and model mismatch, and its transfer function satisfies , where a is the initial amplification factor, ξ is the damping ratio, and τ is the damping ratio. c To compensate for the time constant, s is the Laplace operator, and all parameters satisfy a>0, τ>0, and 0<ξ≤1 to ensure the stability and feasibility of the compensation process; and τ is determined based on the inertial time constants T1, T2, and T3 and the order of the controlled object G(s). c The range of values ​​of F2(s) is used to provide a preset phase lead in the dominant frequency band of the controlled object. At the same time, a and ξ are tuned according to the closed-loop overshoot and adjustment time constraints, so as to improve the dynamic response and anti-disturbance performance of the outlet NOx concentration control without introducing high-frequency amplification effect.

[0034] Tracking controller G c1 (s) and correction controller G c2 (s) all adopt a PI controller structure, and their transfer functions all satisfy K pFor proportionality coefficient, T i Let be the integration time, s be the Laplace operator; and where G is the tracking controller. c1 (s) is used to generate the tracking control quantity for the outlet NOx concentration setpoint r, and its input is the difference between the setpoint processed by F1(s) and the equivalent feedback signal generated by the prediction compensation structure; correction controller G c2 (s) is used to generate the corrective control quantity for model mismatch and disturbance, and its input is the estimated error processed by F2(s); the tracking control quantity and the corrective control quantity are superimposed at the control quantity synthesis node to obtain the ammonia injection feedback correction quantity, so as to achieve decoupling of setpoint tracking and disturbance suppression. Furthermore, the correction channel can be configured with anti-integral saturation and dead zone compensation logic, so that the estimated error is still effectively utilized when the valve is saturated or in a small opening dead zone, avoiding the accumulation of correction quantity leading to overshoot in the recovery stage.

[0035] In embodiments of the present invention, such as Figure 4 As shown, the parameter design and tuning of each link in the predictive compensation feedback control framework includes at least the following sub-steps: S501. Determination of benchmark model and target indicators: Based on the SCR dynamic characteristic model, determine the set of control performance indicators, including at least the upper limit of the tracking error of the outlet NOx concentration, the upper limit of the adjustment time, the upper limit of the overshoot, and the upper limit of the valve opening change rate, and determine the load range and typical operating condition data window used for tuning; S502. Reference Filtering Parameter Selection: Based on the sampling period of the outlet NOx concentration and the noise spectrum characteristics, set the filtering time constant τ of the low-pass filter F1(s). f This ensures that the filtered reference signal retains the SCR-dominant dynamics while suppressing high-frequency components, and guarantees that the phase lag introduced by the filtering stage does not exceed a preset threshold. S503. Matching the predicted internal model with the time delay compensation: Constructing the predicted internal model G m (s) and set its delay compensation stage so that G m (s) matches G(s) in gain and inertia parameters within the dominant frequency band, and its pure hysteresis parameter τ m The internal model output is obtained by satisfying the consistency constraint with the pure time delay τ of the object or by correcting it based on historical data, so as to ensure that the internal model output and the actual output can be used to construct the prediction error after being aligned in time. S504. Initial value tuning of the tracking controller: Setting the tracking controller G based on the time constants T1, T2, T3 and gain parameters of G(s). c1 The proportional coefficient and initial value of integral time of (s) ensure that the closed loop meets the basic stability and controllability requirements when no correction controller is involved, and the initial value is used as the starting point for joint tuning. S505. Joint tuning of phase compensation and correction control: Based on the pure time delay of the object and the phase lag caused by inertia in G(s), determine the compensation time constant and damping ratio of the phase compensation element F2(s), and take the channel of the predicted error entering the correction controller after phase compensation as the object to adjust G(s). c2 (s) is tuned to ensure that the prediction error generates a stable correction control quantity; during the joint tuning process, the index set determined in step S501 is used as a constraint to adjust F2(s) and G. c1 (s) and G c2 (s) Iterative adjustments are made until the export NOx tracking and immunity indicators are met; S506. Engineering Constraint Verification and Online Update: Perform closed-loop simulation or playback verification of the tuned parameters under valve saturation and rate of change constraints. If there is control fluctuation, saturation accumulation, or failure to meet the requirements, return to steps S502~S505 for readjustment. During unit operation, periodically update the parameters based on model residuals or performance degradation criteria to adapt to catalyst activity decay and changes in operating conditions.

[0036] It should be noted that in the predictive compensation feedback control framework constructed in this step, the time delay element is eliminated from the closed-loop characteristic equation, significantly improving the control performance of the large time-delay SCR system; the dual-channel predictive and corrective structure achieves decoupling of setpoint tracking and disturbance suppression, and the tracking controller G... c1 (s) Focuses on rapid, overshoot-free tracking of outlet NOx concentration, with calibration controller G c2 (s) focuses on compensating for model mismatch and inlet NOx disturbances; the two work together to avoid performance trade-offs in multi-objective optimization by a single controller; the phase compensation stage F2(s) introduces a leading phase into the prediction error channel to offset the phase lag caused by time delay and inertia, enabling the correction control quantity to respond to dynamic deviations in a timely manner and improving the system's anti-disturbance capability during load changes; the low-pass filter stage F1(s) suppresses high-frequency control components caused by setpoint changes, reducing the wear of the ammonia injection valve and the energy consumption of the control system; the prediction internal model G m The online update mechanism of (s) enables the control framework to track long-term drift of object characteristics, ensuring the persistence of control accuracy.

[0037] SS6. Ammonia injection valve proactive control: Based on the ammonia injection feedforward control quantity and the ammonia injection feedback correction quantity, an ammonia injection valve control command is generated to regulate the outlet NOx concentration. Specifically, the ammonia injection feedforward control quantity and the ammonia injection feedback correction quantity are synthesized and then subjected to execution constraint processing, including at least upper and lower limit constraints on the ammonia injection valve opening and opening change rate constraints. When the constraints are triggered, the control quantity is subjected to anti-integral saturation processing. At the same time, based on the static mapping relationship between the ammonia injection valve opening and the ammonia injection quantity, the synthesized control quantity is converted into the target valve opening, ensuring that the outlet NOx concentration is preemptively regulated within the valve's reachable range.

[0038] It should be noted that this step achieves the organic integration of predictive control and deviation correction by synthesizing feedforward and feedback control quantities. The feedforward control quantity anticipates future operating condition changes based on load forecasting and flue gas parameter prediction, while the feedback correction quantity corrects model mismatch and disturbances in real time based on the prediction compensation framework. The synergistic effect of the two eliminates the limitations of a single control strategy. The execution constraint processing ensures the executability and safety of control commands by limiting both valve opening and rate of change, preventing valve saturation and excessive wear. The anti-integral saturation processing avoids overshoot and prolonged settling time caused by the accumulation of errors in the integrator during saturation, and improves the speed at which the control system recovers from saturation.

[0039] Example 2: SCR Denitrification Control System Based on Embodiment 1 above, Embodiment 2 further provides an SCR denitrification control system, which adopts the above-mentioned SCR denitrification control method based on load prediction feedforward and prediction compensation feedback, including: The load-flue gas parameter prediction module is used to acquire historical operating data of the unit, including at least boiler load and corresponding fuel and air supply, as well as SCR inlet flue gas temperature and flow rate, and to establish a load-flue gas parameter prediction model based on the historical operating data of the unit. The flue gas parameter and denitrification efficiency prediction module is used to collect the current unit load command and its rate of change in real time as model input, predict the SCR inlet flue gas temperature and flow rate and their trend within a preset time window, and determine the denitrification efficiency η corresponding to the predicted flue gas temperature. The theoretical ammonia demand calculation module is used to establish a theoretical ammonia demand calculation model based on the predicted flue gas flow rate and denitrification efficiency η, combined with the real-time collected SCR inlet NOx concentration and ammonia nitrogen molar ratio NSR, to calculate the theoretical ammonia demand that meets the future instantaneous operating conditions and form the ammonia injection feedforward control quantity. The SCR dynamic characteristic modeling module is used to acquire historical operating data of the SCR denitrification system, including at least the opening degree of the ammonia injection valve, the amount of ammonia injected, and the outlet NOx concentration. It establishes a first-order ammonia flow dynamic characteristic sub-model G2(s) with the opening degree of the ammonia injection valve as input and the amount of ammonia injected as output, and a second-order outlet NOx concentration dynamic characteristic sub-model G1(s) with the amount of ammonia injected as input and the outlet NOx concentration as output. G1(s) and G2(s) are connected in series to form a third-order SCR dynamic characteristic model G(s) with time delay. The predictive compensation feedback control module includes a low-pass filter F1(s), a phase compensation filter F2(s), and a tracking controller G. c1 (s), Correction controller G c2 (s), Estimated internal model G m The (s) and its time delay compensation stage take the outlet NOx concentration setpoint r as input, construct the prediction error by the deviation between the internal model output and the actual output, and perform phase compensation on the prediction error through F2(s) to suppress the dynamic mismatch caused by time delay. The phase-compensated prediction error is then input into G. c2 (s) Generate correction control quantity and output ammonia injection feedback correction quantity; The ammonia injection valve advance control module is used to generate ammonia injection valve control commands based on the ammonia injection feedforward control quantity and the ammonia injection feedback correction quantity, so as to adjust the ammonia injection quantity and realize the outlet NOx concentration control.

[0040] Example 3: Application Case Based on Examples 1 and 2 above, Example 3 further provides a practical application example combining a 300MW coal-fired power generating unit's SCR denitrification system to verify the feasibility of the control method provided by this invention and its significant technical effects under load variation conditions. This unit is equipped with a single-furnace subcritical natural circulation boiler, employing a double-layer SCR denitrification reactor with a catalyst loading of 3+1 layers. The designed denitrification efficiency is ≥85%, and the target NOx concentration at the outlet is ≤50 mg / Nm³. 3 The unit participates in deep peak shaving operation, with load variations ranging from 40% to 100% of rated load.

[0041] Historical operating data of the unit over the past 6 months was collected. After time alignment, anomaly removal, and normalization, a load-flue gas parameter prediction model based on an LSTM network was constructed. The network structure consisted of three hidden layers (with 128, 64, and 32 neurons respectively), with a Dropout rate of 0.2. Training was performed using the Adam optimizer with a learning rate of 0.001. After training, the model achieved a flue gas temperature prediction error of <±3℃ and a flue gas flow prediction error of <±2% on the test set, with the prediction time window set to the next 10 minutes.

[0042] The corresponding η is obtained based on the pre-calibrated temperature T-denitrification efficiency η mapping relationship, and combined with the real-time collected SCR system inlet NOx concentration C. NOx Calculating the ammonia feedforward demand U based on the ammonia-nitrogen molar ratio (NSR) ff Specifically Taking a certain period of historical operating data as an example, the flue gas flow rate Q = 2.0 × 10⁻⁶. 6 Nm 3 / h, CNOx=300 mg / Nm 3 η=83.3%, ammonia injection feedforward control quantity U ff =2.0×10 6 ×300×0.833 / (0.9×10 6 )×17 / 14≈205.2 kg / h.

[0043] Based on operating data collected under steady-state conditions, a step excitation of 3% amplitude and 180 seconds was applied to the ammonia injection valve, and the dynamic response of ammonia injection rate and outlet NOx concentration was recorded. The recursive least squares method was used to identify model parameters, obtaining a sub-model of the dynamic characteristics of outlet NOx concentration. Ammonia flow dynamic characteristic sub-model , .

[0044] Based on this model, the parameters of each link in the predicted compensation feedback control framework are as follows: Low-pass filter link F1(s) = 1 / (60s+1), τ f =60 seconds; the phase compensation stage adopts a second-order structure, with a=3 and ξ=0.5, corresponding to an attenuation rate of 0.973, according to , The calculated time is τ≈14 seconds, and the equivalent inertial time after compensation is T'≈68 seconds. Tracking controller G c1 (s) = 0.08(1 + 1 / (80s)), proportionality constant K p1 =0.08, integration time T i1 =80 seconds; Correction controller G c2 (s) = 0.08(1 + 1 / (280s)), proportionality constant K p2 =0.08, integration time T i2 =280 seconds; Estimated internal model G m G(s) has the same structure as G(s), with parameters updated synchronously and a time delay τ. m =111 seconds.

[0045] The feedforward control quantity U ff With feedback correction amount U fbAfter synthesis, the valve opening and rate of change are constrained and converted into a target valve opening command. The control command is transmitted to the ammonia injection regulating valve actuator through the PLC's analog output module (4-20mA signal) to achieve closed-loop regulation of the outlet NOx concentration. When the valve triggers saturation, the conditional integral anti-saturation logic is activated to prevent integrator accumulation.

[0046] To verify the technical advantages of the method of this invention, comparative tests were conducted using operating data of the unit under variable load conditions. The operating conditions were set as follows: the load stepped to 75% and stabilized between t=0 and 2500 seconds, and the load stepped to 50% and stabilized between t=2500 and 3500 seconds, with a load change rate of approximately 2% of rated load per minute. The outlet NOx concentration setpoint was r=50 mg / Nm³. 3 The comparative solutions include: improved Smith predictor compensation control (the method of this invention), conventional Smith predictor control (without phase compensation and dual-channel structure), and conventional cascade PID control.

[0047] Table 1 below presents a performance comparison of different control schemes under varying load conditions. As shown in Table 1, during the load increase phase (t=1000s), the peak NOx concentration at the outlet reached 5.7 mg / Nm³ with an overshoot of 14% and a settling time of 450 seconds under conventional cascade PID control; the peak concentration under conventional Smith predictor control was 5.5 mg / Nm³ with an overshoot of 10% and a settling time of 320 seconds; while the method of this invention achieved a peak concentration of only 5.1 mg / Nm³ with an overshoot of only 2% and a settling time reduced to 180 seconds. Compared to conventional cascade PID control, the method of this invention reduced the overshoot by 85.7% and shortened the settling time by 60%; compared to conventional Smith predictor control, the overshoot was reduced by 80% and the settling time was shortened by 43.8%.

[0048] During the load reduction phase (t=2500s), the conventional cascade PID control exhibited severe overshoot, with the outlet NOx concentration dropping to a minimum of 0.5 mg / Nm³. 3 This was followed by violent oscillations, with a settling time exceeding 600 seconds; the lowest value controlled by the conventional Smith predictor was 3.4 mg / Nm³. 3 Slight oscillations occurred, with an adjustment time of approximately 400 seconds; the method of this invention, however, maintained a smooth transition, with the lowest NOx concentration maintained at 4.8 mg / Nm³. 3 There was no obvious oscillation, and the adjustment time was only 200 seconds.

[0049] In terms of steady-state control performance, the outlet NOx concentration fluctuation amplitude of the method of this invention is only ±0.3 mg / Nm³. 3 The conventional Smith predictor control is ±0.6 mg / Nm³. 3Conventional cascade PID control achieves a steady-state ripple amplitude of ±0.8 mg / Nm³. Compared to conventional cascade PID control, the method of this invention reduces the steady-state ripple amplitude by 62.5%; compared to conventional Smith predictor control, it reduces it by 50%.

[0050] Table 1. Performance Comparison of Different Control Schemes under Variable Load Conditions The objectives of this invention have been fully and effectively achieved through the above embodiments. Those skilled in the art will understand that this invention includes, but is not limited to, the contents described in the accompanying drawings and the specific embodiments described above. Although the invention has been described with reference to what is currently considered the most practical and preferred embodiments, it should be understood that the invention is not limited to the disclosed embodiments, and any modifications that do not depart from the functional and structural principles of the invention will be included within the scope of the claims.

Claims

1. A method for SCR denitrification control based on load prediction feedforward and prediction compensation feedback, characterized in that, Includes the following steps: SS1. Acquire historical operating data of the unit, including boiler load and corresponding fuel and air supply, as well as SCR inlet flue gas temperature and flow rate, and establish a load-flue gas parameter prediction model based on the historical operating data of the unit; SS2. Real-time acquisition of current unit load commands and their rate of change, used as input variables for load-flue gas parameter prediction model, to predict the SCR inlet flue gas temperature and flow rate and their trend within a preset time window, and to determine the denitrification efficiency η corresponding to the predicted flue gas temperature; SS3. Based on the predicted flue gas flow rate and the denitrification efficiency η corresponding to the predicted flue gas temperature, combined with the real-time collected SCR inlet NOx concentration and ammonia nitrogen molar ratio NSR, a theoretical ammonia demand calculation model is established to calculate the theoretical ammonia demand that meets the future instantaneous operating conditions, and to form the ammonia injection feedforward control quantity. SS4. Obtain historical operating data of the SCR denitrification system, including ammonia injection valve opening, ammonia injection rate, and outlet NOx concentration; Based on the historical operating data of the SCR denitrification system, parameter identification was performed, and a first-order ammonia flow dynamic characteristic sub-model with ammonia injection valve opening as input and ammonia injection quantity as output was established. A second-order outlet NOx concentration dynamic characteristic sub-model with time delay was established with ammonia injection quantity as input and outlet NOx concentration as output. The two sub-models were connected in series to form a third-order SCR dynamic characteristic model G(s) with time delay. SS5. Design a predictive compensation feedback control framework based on G(s), including at least a low-pass filter F1(s), a phase compensation circuit F2(s), and a tracking controller G. c1 (s), Correction controller G c2 (s), Estimated internal model G m (s) and its time delay compensation process, G m (s) matches G(s) in gain and inertia parameters within the dominant frequency band, and G m (s) Pure time delay parameter τ of the time delay compensation stage m The pure time delay parameter τ of the object with respect to G(s) satisfies the consistency constraint; the phase compensation element F2(s) is based on G(s) and G... m The phase difference between G(s) is designed to provide a preset phase lead within the dominant frequency band of G(s); and, Using the outlet NOx concentration setpoint r as input, r is processed by F1(s) to obtain a smoothed reference signal; G c1 (s) Output tracking control quantity u1, one input G of u1 m (s) Obtain the estimated intermediate quantity P m (s), P m (s) The internal model output y is obtained after the time delay compensation process. m (t), output y through the internal model m The deviation between (t) and the actual output y(t) of the SCR denitrification system constitutes the prediction error e. m (t), and use F2(s) to evaluate the prediction error e. m (t) Implement phase compensation to suppress dynamic mismatch caused by time delay, and reduce the prediction error e after phase compensation. m (t) Input G c2 (s) Generate the correction control quantity u2, and at the control quantity synthesis node, output the ammonia injection feedback correction quantity after subtracting the correction control quantity u2 from the tracking control quantity u1; and where e m (t) and P m (s) is superimposed at the main bias construction node, and then compared with the smoothed reference signal to form the main bias signal, which serves as the tracking controller G. c1 Input of (s); SS6. Generate ammonia injection valve control commands based on the ammonia injection feedforward control quantity and the ammonia injection feedback correction quantity.

2. The method according to claim 1, characterized in that, In step SS1, the load-flue gas parameter prediction model is a multi-input multi-output prediction model. It takes boiler load, fuel supply and air supply as input variables, and SCR inlet flue gas temperature and flow rate as output variables. It also outputs the changing trend of SCR inlet flue gas temperature and flow rate, and introduces at least one load-related dynamic characteristic as an auxiliary input. The dynamic characteristic is the load change rate and the fuel or air supply change rate. The load-flue gas parameter prediction model adopts a time-series prediction architecture based on a Long Short-Term Memory (LSTM) network. This includes: using boiler load, corresponding fuel and air supply, and dynamic features as input feature vectors for the LSTM network, and SCR inlet flue gas temperature and flow rate as output target variables; extracting the time-series evolution features of flue gas parameters through hidden layers, training the LSTM network weight parameters using backpropagation, and utilizing a Dropout layer to prevent overfitting, thus converging the prediction error to a preset threshold range; and predicting the dynamic change trend of SCR inlet flue gas temperature and flow rate within the next 5-15 minutes based on current and historical load data.

3. The method according to claim 1, characterized in that, In step SS2, the method for determining the denitrification efficiency η is as follows: a mapping relationship database between flue gas temperature T and denitrification efficiency η is established in advance through SCR reactor catalyst activity test experiments. The mapping relationship is represented as a piecewise function. When T < 300℃, η increases exponentially with increasing temperature. When 300℃ ≤ T ≤ 420℃, η is maintained in the high-efficiency range of 0.80~0.

95. When T > 420℃, η decreases due to the intensified ammonia oxidation reaction. After predicting the flue gas temperature in real time, the instantaneous denitrification efficiency η value under the corresponding operating condition is obtained by querying the mapping relationship database and performing linear interpolation calculation. When the predicted flue gas temperature exceeds the calibrated temperature range, η is subjected to boundary truncation or extrapolation constraint processing.

4. The method according to claim 1, characterized in that, In step SS3, the theoretical ammonia demand calculation model is based on the flue gas flow rate Q and the inlet NOx concentration C. NOx The denitrification efficiency η and the ammonia-to-nitrogen molar ratio NSR are used to construct a material balance relationship, and the molar mass ratio M of ammonia to NOx is introduced for unit conversion, so that the ammonia injection feedforward control quantity U is... ff Satisfy U ff =Q×C NOx ×η×M / NSR, and in the calculation process, the flue gas flow rate Q is converted from volume basis to molar flow rate or mass flow rate, NOx is converted to a uniform basis oxygen content or uniform dry basis conditions, and C is adjusted accordingly. NOx Moving average filtering is used to suppress measurement noise.

5. The method according to claim 1, characterized in that, In step SS4, parameter identification is performed based on historical operating data of the SCR denitrification system to construct a sub-model G1(s) of the dynamic characteristics of the outlet NOx concentration and a sub-model G2(s) of the dynamic characteristics of the ammonia flow rate, including the following sub-steps: S411. Acquisition of historical operating data: Conduct identification tests on the SCR denitrification system or collect historical operating data of the SCR denitrification system using available identification data windows formed during normal unit operation, including at least the opening degree of ammonia injection valve, ammonia injection rate, and outlet NOx concentration, and perform preprocessing on the data; S412. Identification of the Dynamic Characteristics Sub-model for Outlet NOx: Using ammonia injection rate as input and outlet NOx concentration as output, parameter identification is performed to establish a dynamic characteristics sub-model for outlet NOx concentration. Its transfer function is expressed using a second-order inertial element combined with a pure time delay structure, satisfying... Where K1 is the gain parameter, T1 and T2 are time constants, τ is the object pure time delay parameter of G(s), used to characterize the inertia and pure time delay characteristics of the response from the change in ammonia injection rate to the outlet NOx, and s is the Laplace operator. Indicates the delay stage; S413. Identification of the dynamic characteristic sub-model of ammonia flow rate: Using the opening degree of the ammonia injection valve as the input and the ammonia injection rate as the output, the transfer function of the dynamic characteristic sub-model of ammonia flow rate is established. Where K2 is the gain parameter, T3 is the time constant, and s is the Laplace operator; S414. Construction of SCR Dynamic Characteristic Model: The transfer functions G1(s) and G2(s) of the two sub-models are connected in series to form the SCR dynamic characteristic model. G0(s) is an SCR object used to characterize the overall dynamic transmission process from the change in the opening degree of the ammonia injection valve, through the amount of ammonia injected, to the change in the NOx concentration at the outlet, and serves as the object model for the subsequent prediction, compensation, and feedback control framework.

6. The method according to claim 5, characterized in that, In step SS4, the parameter identification employs recursive least squares or maximum likelihood methods, combined with a forgetting factor mechanism to achieve online updates. Before identification, the input and output data are preprocessed, and persistent excitation constraints are applied to the input and output data, including: applying a step excitation of preset amplitude and duration to the ammonia injection valve opening or ammonia injection quantity, ensuring sufficient dynamic change amplitude and frequency band coverage within the identification time window; performing correlation analysis and impulse / step response analysis on historical data of ammonia injection quantity and outlet NOx concentration, determining the parameters of G1(s) based on the dynamic response of outlet NOx to ammonia injection quantity; performing time-series correlation and step response analysis on historical data of ammonia injection valve opening and ammonia injection quantity, determining the parameters of G2(s) based on the response curve of ammonia injection quantity to valve opening; the forgetting factor λ is limited to (0,1), and adaptively adjusted based on model residuals and operating condition change rates, ensuring that the parameters of G1(s) and G2(s) are recursively updated during operation.

7. The method according to claim 5, characterized in that, In step SS5, the prediction compensation feedback control framework adopts an improved Smith predictor architecture, the topology of which is as follows: Using the outlet NOx concentration setpoint r as input, r is processed by a low-pass filter F1(s) to obtain a smoothed reference signal; based on the tracking controller G... c1 The output of (s) constructs a dual-channel prediction and correction structure, where: G c1 (s) One output u1 is connected to the correction controller G at the control input synthesis node. c2 The output of (s) is algebraically synthesized to obtain the SCR object driving signal U(s), and U(s) is combined with the inlet NOx disturbance D. NOx (s) is superimposed at the input of the SCR object and used as the input of the SCR object G0(s), which then passes through the object pure time delay stage of G0(s) and G(s). After processing, the outlet NOx concentration output y(t) is obtained; G c1 (s) Output u1's other input to predict the internal model G m (s) Obtain the estimated intermediate quantity P m (s), and P m (s) After the internal model pure lag stage Obtain the internal model output y m (t), where τ m To estimate the internal model G m (s) The pure time delay parameter of the delay compensation stage; y(t) and y m (t) The difference at the comparison node yields the estimated error e. m (t)=y(t)-y m (t); the estimated error e m (t) is processed by the phase compensation stage F2(s) and then input to the correction controller G. c2 (s), outputting the correction control quantity u2 and participating in the synthesis of the object driving signal U(s); simultaneously, the predicted error e m (t) and the estimated intermediate value P m (s) The main bias signal is superimposed at the main bias construction node and compared with the smoothed reference signal to form the main bias signal, which serves as the tracking controller G. c1 The input of (s) forms a predictive feedback closed loop with phase compensation.

8. The method according to claim 1, characterized in that, In step SS5, the low-pass filter F1(s) is set on the input side of the outlet NOx concentration setpoint r or on the differential side between the setpoint and the feedback value, and F1(s) is a first-order inertial low-pass filter whose transfer function satisfies , where τ f Let F1(s) be the filtering time constant, s be the Laplace operator, and the cutoff frequency of F1(s) be set based on the sampling period of the outlet NOx concentration and the noise spectrum characteristics of the measurement signal; furthermore, The phase compensation stage F2(s) is set on the prediction error channel formed by the internal model output and the actual output, and F2(s) adopts a second-order phase compensation network structure, based on G(s) and the predicted internal model G. m The phase difference between (s) is designed to compensate for the phase lag caused by time delay and model mismatch, and its transfer function satisfies Where a is the initial amplification factor, ξ is the damping ratio, and τ is the damping ratio. c To compensate for the time constant, s is the Laplace operator, and all parameters satisfy a>0, τ c >0, 0<ξ≤1; and determine τ based on the time constants T1, T2, T3 and order of the controlled object's SCR dynamic characteristic model G(s). c The range of values ​​of F2(s) is used to provide a preset phase lead within the dominant frequency band of the SCR dynamic characteristic model G(s) of the controlled object, while a and ξ are tuned according to the closed-loop overshoot and adjustment time constraints.

9. The method according to claim 8, characterized in that, In step SS5, the tracking controller G c1 (s) and correction controller G c2 (s) all adopt a PI controller structure, satisfying K p For proportionality coefficient, T i Let be the integration time, s be the Laplace operator; and where G is the tracking controller. c1 (s) is used to generate the tracking control quantity for the outlet NOx concentration setpoint r. Its input is the difference between the setpoint processed by F1(s) and the equivalent feedback signal generated by the predictive compensation feedback control framework. This difference forms the main deviation signal; the correction controller G c2 F2(s) is used to form the corrective control quantity for model mismatch and disturbances, and its input is the predicted error processed by F2(s).

10. The method according to claim 9, characterized in that, In step SS5, the parameter design and tuning of each link in the predictive compensation feedback control framework includes at least the following sub-steps: S501. Determination of benchmark model and target indicators: Based on the SCR dynamic characteristic model, determine the set of control performance indicators, including at least the upper limit of the tracking error of the outlet NOx concentration, the upper limit of the adjustment time, the upper limit of the closed-loop overshoot, and the upper limit of the valve opening change rate, and determine the load range and typical operating condition data window used for tuning; S502. Reference Filtering Parameter Selection: Based on the sampling period of the outlet NOx concentration and the noise spectrum characteristics, set the filtering time constant τ of the low-pass filter F1(s). f ; S503. Matching the predicted internal model with the time delay compensation: Constructing the predicted internal model G m (s) and set its delay compensation stage so that G m (s) matches G(s) in gain and inertia parameters within the dominant frequency band, and G m (s) Pure time delay parameter τ of the time delay compensation stage m The pure time delay parameter τ of the object with respect to G(s) satisfies the consistency constraint or is obtained by correction based on historical data; S504. Initial value tuning of the tracking controller: Setting the tracking controller G based on the time constants T1, T2, T3 and gain parameters of G(s). c1 The proportional coefficient (s) and the initial value of the integration time are used as the starting point for joint tuning; S505. Joint tuning of phase compensation and correction control: Based on the pure time delay of the object and the phase lag caused by inertia in G(s), determine the compensation time constant and damping ratio of the phase compensation element F2(s), and take the channel of the predicted error entering the correction controller after phase compensation as the object to adjust G(s). c2 (s) is tuned; during the joint tuning process, the set of indices determined in step S501 is used as constraints to adjust F2(s) and G. c1 (s) and G c2 (s) performs iterative adjustments; S506. Engineering constraint verification and online update: Perform closed-loop simulation or playback verification of the tuned parameters under valve saturation and rate of change constraints. If there is control quantity jitter, saturation accumulation or failure to meet the index, return to steps S502~S505 for readjustment.

11. The method according to claim 1, characterized in that, In step SS6, the ammonia injection feedforward control quantity and the ammonia injection feedback correction quantity are combined and then subjected to execution constraint processing, including upper limit and lower limit constraints of the ammonia injection valve opening and opening change rate constraints, and the control quantity is subjected to anti-integral saturation processing when the constraint is triggered. Based on the static mapping relationship between the ammonia injection valve opening and the ammonia injection quantity, the synthesized control quantity is converted into the valve target opening.

12. An SCR denitrification control system, characterized in that, Includes the following modules: The load-flue gas parameter prediction module is used to acquire historical operating data of the unit, including at least boiler load and corresponding fuel and air supply, as well as SCR inlet flue gas temperature and flow rate, and to establish a load-flue gas parameter prediction model based on the historical operating data of the unit. The flue gas parameter and denitrification efficiency prediction module is used to collect the current unit load command and its rate of change in real time as model input, predict the SCR inlet flue gas temperature and flow rate and their trend within a preset time window, and determine the denitrification efficiency η corresponding to the predicted flue gas temperature. The theoretical ammonia demand calculation module is used to establish a theoretical ammonia demand calculation model based on the predicted flue gas flow rate and denitrification efficiency η, combined with the real-time collected SCR inlet NOx concentration and ammonia nitrogen molar ratio NSR, to calculate the theoretical ammonia demand that meets the future instantaneous operating conditions and form the ammonia injection feedforward control quantity. The SCR dynamic characteristic modeling module is used to acquire historical operating data of the SCR denitrification system, including at least the opening degree of the ammonia injection valve, the amount of ammonia injected, and the outlet NOx concentration. It establishes a first-order ammonia flow dynamic characteristic sub-model G2(s) with the opening degree of the ammonia injection valve as input and the amount of ammonia injected as output, and a second-order outlet NOx concentration dynamic characteristic sub-model G1(s) with the amount of ammonia injected as input and the outlet NOx concentration as output. G1(s) and G2(s) are connected in series to form a third-order SCR dynamic characteristic model G(s) with time delay. The predictive compensation feedback control module includes a low-pass filter F1(s), a phase compensation circuit F2(s), and a tracking controller G. c1 (s), Correction controller G c2 (s), Estimated internal model G m (s) and its time delay compensation process, G m (s) matches G(s) in gain and inertia parameters within the dominant frequency band, and G m (s) Pure time delay parameter τ of the time delay compensation stage m The pure time delay parameter τ of the object with respect to G(s) satisfies the consistency constraint; the phase compensation element F2(s) is based on G(s) and G... m The phase difference between G(s) is designed to provide a preset phase lead within the dominant frequency band of G(s); and, Using the outlet NOx concentration setpoint r as input, r is processed by F1(s) to obtain a smoothed reference signal; G c1 (s) Output tracking control quantity u1, one input G of u1 m (s) Obtain the estimated intermediate quantity P m (s), P m (s) The internal model output y is obtained after the time delay compensation process. m (t), output y through the internal model m The deviation between (t) and the actual output y(t) of the SCR denitrification system constitutes the prediction error e. m (t), and use F2(s) to evaluate the prediction error e. m (t) Implement phase compensation to reduce the prediction error e after phase compensation. m (t) Input G c2 (s) Generate the correction control quantity u2, and at the control quantity synthesis node, output the ammonia injection feedback correction quantity after subtracting the correction control quantity u2 from the tracking control quantity u1; and where e m (t) and P m (s) is superimposed at the main bias construction node, and then compared with the smoothed reference signal to form the main bias signal, which serves as G. c1 Input of (s); The ammonia injection valve advance control module is used to generate ammonia injection valve control commands based on the ammonia injection feedforward control quantity and the ammonia injection feedback correction quantity, adjust the ammonia injection quantity, and achieve outlet NOx concentration control.

13. A computer program product comprising computer instructions, characterized in that, The computer instructions are used to execute the steps of the SCR denitrification control method based on load prediction feedforward and prediction compensation feedback as described in any one of claims 1 to 11.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the SCR denitrification control method based on load prediction feedforward and prediction compensation feedback as described in any one of claims 1 to 11.

Images