A denitration method and device suitable for wide load operation of a unit

Abstract

This invention relates to a denitrification method and equipment adapted to wide-load operation of a generating unit. The equipment includes: a full-section multi-point sampling module with M×N gridded measuring points at the SCR inlet and outlet flues; an integrated ammonia-nitrogen-temperature auxiliary module installed at an intermediate elevation between the two reactor sides, integrating dust removal components and preheating backflushing interfaces; an integrated ammonia-nitrogen-temperature measurement module; a decoupling and coupling control module to realize uniform measurement, alternating measurement, alternating interlocking, and backflushing cleaning modes; and an automatic ammonia injection control module using a fanless, low-power controller. The method includes: multimodal homogeneous sampling (uniform measurement and alternating measurement interlocking); constructing a deep neural network to predict future NOx and ammonia escape, with online incremental updates; multi-objective constraint optimization feedforward-feedback control, with feedback coefficients tuned in real time based on model gradients; and switching to the feedforward-dominant mode when the load change rate exceeds a threshold. The advantages of this invention are: improved denitrification control accuracy under wide loads and reduced ammonia consumption and ammonia escape risk.

Description

Technical Field

[0001] This invention relates to a denitrification method and equipment adapted to wide-load operation of power units, belonging to the field of flue gas denitrification in coal-fired power plants. Background Technology

[0002] Currently, coal-fired power plants commonly employ Selective Catalytic Reduction (SCR) denitrification technology, which reduces nitrogen oxides (NOx) in flue gas to nitrogen and water under the action of a catalyst to achieve ultra-low emissions. The control of SCR denitrification systems typically relies on continuous flue gas monitoring systems (CEMS) installed at the flue gas inlet and outlet to provide data on NOx, O2, and ammonia slip, and adjusts the ammonia injection rate based on PID control logic. Most units have installed online monitoring equipment at the inlet and outlet of the denitrification reactor, and the main ammonia injection valve is automatically controlled, while the branch ammonia injection grilles (AIGs) rely on manual adjustment.

[0003] However, with the large-scale grid connection of new energy sources, coal-fired power units need to frequently participate in deep peak shaving and operate under wide load (40%~100% rated load) or even extremely low load conditions for a long time. Under these conditions, flue gas volume, flue gas temperature, NOx concentration and flow field distribution change drastically. Existing technologies have exposed the following pain points: (1) Monitoring is lagging and has poor representativeness: Traditional CEMS is usually installed at the bend of the flue or at a short straight section. The sampling point is single and cannot represent the flue gas composition of the whole section. The analysis delay is as long as several minutes, which leads to serious lag in ammonia injection control; (2) Excessive ammonia injection is prominent: In order to cope with the instantaneous NOx exceedance at the outlet caused by load changes, operators are forced to inject excessive ammonia, which leads to a chain of problems such as catalyst poisoning, air preheater ammonium bisulfate blockage, ammonia escape exceeding the standard, dust collector blockage and increased fan power consumption; (3) Single control mode: Conventional PID control is difficult to cope with the large inertia, large delay, strong nonlinearity and multivariable coupling characteristics of SCR system. Especially when the load changes rapidly, it is impossible to achieve fine and feedforward ammonia injection control; (4) Lack of zonal diagnosis capability: It is impossible to know the ammonia nitrogen concentration distribution in different areas inside the flue, which leads to the lack of scientific basis for AIG manual valve adjustment. The uneven distribution of flow field and concentration field further aggravates the decline in denitrification efficiency. Summary of the Invention

[0004] To address the problems of poor monitoring representativeness, delayed response, excessive ammonia injection, single control mode, and lack of zonal diagnostics in the existing technologies, this invention provides a denitrification method and equipment suitable for wide-load operation of the unit. By constructing a full-section, homogeneous, and rapid real-time ammonia nitrogen and temperature monitoring system, and introducing a data-driven adaptive prediction model and decoupled-coupled control logic, accurate, rapid, and multi-objective optimized control of ammonia injection under wide-load conditions is achieved. This significantly reduces the amount of ammonia injected, minimizes ammonia slip, extends the lifespan of auxiliary materials, and improves the stability and safety of unit operation while ensuring environmental compliance.

[0005] This invention provides a denitrification method and equipment adapted to wide-load operation of power units. The technical solution of this invention is: A denitrification device adapted to wide-load operation of a generating unit, comprising: A full-section multi-point sampling module includes several sampling probes arranged in a matrix on the cross-section of the SCR reactor inlet flue and outlet flue, which are used to collect flue gas samples from different zones. An integrated ammonia nitrogen and temperature auxiliary module has its air inlet connected to several sampling probes via pipelines equipped with programmable sampling valves, for collecting, removing dust, and mixing the flue gas samples. An integrated ammonia nitrogen and temperature measurement module is fixedly installed on the detection cell interface of the integrated ammonia nitrogen and temperature auxiliary module. It is used to measure the concentrations of NO, NO2, NH3, and O2 in the mixed flue gas, as well as the temperature, pressure, and flow rate, and to generate real-time monitoring signals. The decoupling and coupling control module includes several programmable sampling valves, several programmable backflush valves, and a control panel. The control panel is electrically connected to each programmable sampling valve and each programmable backflush valve to control the on / off state, so as to realize the uniform measurement coupling mode, the zoned rotation measurement decoupling mode, the alternating interlocking mode, and the backflush cleaning mode for flue gas. The ammonia injection automatic control module is connected to the integrated ammonia-nitrogen-temperature measurement module and the unit's DCS system. It is used to output optimized ammonia injection quantity control commands and zoned ammonia injection adjustment suggestions based on the real-time monitoring signals and unit operating parameters.

[0006] The full-section multi-point sampling module sets up a measurement section at both the inlet and outlet flues of the SCR reactor. Each measurement section has 3 or 4 measurement points in the width direction and 3 or 4 measurement points in the depth direction, forming an M×N gridded measurement point matrix, where M is 3 or 4 and N is 3 or 4. Furthermore, the ammonia nitrogen temperature integrated auxiliary module is installed at the intermediate elevation between the two independent reactor sides of the SCR reactor.

[0007] The integrated ammonia nitrogen and temperature auxiliary module integrates a high-temperature flue gas cyclone dust removal or gravity settling dust removal component, and is equipped with a compressed air preheating backflush interface, which is connected to the backflush valve in the decoupling and coupling control module.

[0008] The ammonia injection automatic control module adopts a fanless, low-power industrial controller, supports the Modbus RTU communication protocol, and achieves redundant and seamless switching connection with the unit's DCS system. When the ammonia injection automatic control module fails, the control of the ammonia injection regulating valve is automatically and seamlessly returned to the DCS native control system.

[0009] A denitrification method based on the aforementioned denitrification equipment adapted to wide-load operation of the unit includes the following steps: S1. Multimodal homogeneous sampling and real-time monitoring: First, open the sampling valves of all zones, mix the flue gas of the entire cross section and perform integrated measurement to obtain the average parameters of the entire cross section; then, according to the preset time period, close the sampling valves of other zones in sequence and open only a single zone to perform round-robin measurement on the zone to obtain the zone characteristic parameters. S2. Construct and run a wide-load adaptive prediction model: Using unit load, inlet NOx concentration, flue gas temperature, flue gas flow rate, historical ammonia injection rate, and historical ammonia slip concentration as input features, establish and train a deep neural network model to predict the outlet NOx concentration and ammonia slip concentration for the next 3-10 minutes; the deep neural network model is updated online incrementally based on actual measurement data during unit operation; S3. Perform multi-objective constraint optimization control: With the environmental NOx emission limit as a hard constraint, and minimizing the ammonia injection amount and maximizing the denitrification efficiency as dual optimization objectives, the model prediction value from step S2 is used for feedforward control, and combined with feedback PID control based on model gradient information to tune parameters in real time, the total ammonia injection amount command is calculated together; at the same time, based on the characteristic parameters of each zone obtained in step S1, the adjustment suggestions for each branch of the ammonia injection grid are calculated and output. S4. Wide-load condition adaptive switching and execution: Real-time monitoring of unit load change rate. When the load change rate exceeds the preset threshold, the control strategy is automatically switched to feedforward dominant mode to reduce the weight of feedback control. When the load returns to stability, it switches back to normal control mode and sends the calculated ammonia injection command to be executed through the control system.

[0010] The deep neural network model in step S2 is a BP neural network with two hidden layers; its input layer feature vector Specifically, this includes: real-time load of the generating unit. Average NOx concentration at SCR inlet Average temperature of flue gas inside the reactor flue gas flow rate Total ammonia injection volume at the previous moment The average ammonia escape concentration at the export point at the previous moment Output layer prediction vector The predicted NOx concentration at the outlet at time τ in the future. Predicted ammonia escape concentration The model was trained using the mean squared error loss function and the Adam optimizer, with an initial learning rate of 0.001 and 500 training epochs.

[0011] In step S3, the total ammonia injection command is calculated. The formula is: Among them, feedforward quantity Kstoich represents the stoichiometric correction coefficient identified online by the adaptive prediction model, which is dimensionless; NO xin The average NOx concentration at the SCR reactor inlet is expressed in mg / Nm³; FlueGas represents the flue gas flow rate in Nm³ / h; MNO x MNH3 represents the molar mass of nitrogen oxides; SPNO represents the molar mass of ammonia. x This represents the dynamically optimized NOx target value at the outlet, expressed in mg / Nm³. The feedback correction ammonia injection quantity The calculation formula is: ;in, , , These represent the proportional coefficient, integral coefficient, and differential coefficient, respectively, which change with time t. The sensitivity gradient of the ammonia injection rate is based on the output of the adaptive prediction model. Perform real-time adjustment; This represents the deviation of the outlet NOx concentration at time t, expressed in mg / Nm³, and is calculated using the following formula: ; This represents the measured NOx concentration at the SCR reactor outlet at time t, in mg / Nm³.

[0012] Step S3 further includes: when the ammonia escape prediction value output by the adaptive prediction model... Or measured ammonia slip value When the ammonia escape rate exceeds 3 ppm, the protection function is triggered, blocking any further increase in the ammonia injection regulating valve and forcibly reducing the total ammonia injection rate by 5% until the ammonia escape value drops below 2.5 ppm.

[0013] The uniform measurement mode and the rotation measurement mode in step S1 are executed in an interleaved manner. The duration of the uniform measurement mode is 50 to 70 minutes, and the duration of the rotation measurement mode for each single zone is 8 to 12 minutes. Furthermore, the backflushing cleaning mode is triggered by a timer or by the backflushing request signal from the integrated measurement module. During the backflushing, all programmable sampling valves are closed, and the programmable backflushing valves are opened to perform pulse cleaning on the sampling pipeline and the integrated auxiliary module. At the same time, the measurement module maintains the previous value output.

[0014] The preset threshold in step S4 is set to an absolute value of the load change rate greater than 3% / minute; in the feedforward-dominated mode, the feedforward quantity... In general instructions The computational weight in the process is increased to 80% to 100%, and the integral term of the feedback PID controller is frozen.

[0015] The advantages of this invention are: 1. By combining full-section multi-point sampling with the ammonia nitrogen temperature integrated auxiliary module for co-source processing, the measured values ​​truly reflect the flue gas composition across the entire flue section, significantly reducing the representativeness bias of traditional single-point sampling; the use of in-situ ultraviolet differential absorption spectroscopy measurement significantly shortens the system response time, providing a real-time and reliable data foundation for feedforward control.

[0016] 2. By combining an adaptive prediction model based on neural networks with a multi-objective optimization control algorithm, the average ammonia injection rate can be effectively reduced under wide load variation conditions, and the outlet ammonia escape can be stably controlled within the ideal range, preventing ammonium bisulfate blockage and catalyst poisoning in the air preheater.

[0017] 3. The NOx concentration at the outlet can be stably controlled within the environmental protection limit under both steady-state and rapid load change conditions, eliminating the risk of exceeding the standard and reducing the operating burden on operators.

[0018] 4. Avoid frequent blockages and forced shutdowns for cleaning of the air preheater and dust collector caused by excessive ammonia injection, extend the service life of the catalyst, and reduce the power consumption and maintenance costs of the induced draft fan.

[0019] 5. It has ammonia escape protection logic and a seamless switching function with DCS to ensure that the system safety is no less than that of the original DCS control system, and optimizes the automatic seamless return to the original DCS control when the system fails. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the device of the present invention.

[0021] Figure 2 This is a flowchart illustrating the method of the present invention.

[0022] Figure 3 This is a diagram showing the layout of the measurement points for the full-section multi-point sampling module of the present invention. Detailed Implementation

[0023] The present invention will be further described below with reference to specific embodiments, and the advantages and features of the present invention will become clearer as a result. However, these embodiments are merely exemplary and do not constitute any limitation on the scope of the present invention. Those skilled in the art should understand that modifications or substitutions can be made to the details and form of the technical solutions of the present invention without departing from the spirit and scope of the present invention, but all such modifications and substitutions fall within the protection scope of the present invention.

[0024] See Figures 1 to 3 This invention relates to a denitrification device adapted to wide-load operation of a generating unit, comprising: The full-section multi-point sampling module 1 includes several sampling probes arranged in the inlet flue and outlet flue of the SCR reactor. The sampling probes are distributed in a matrix on the cross-section of the flue and are used to collect flue gas samples from different zones. The ammonia nitrogen temperature integrated auxiliary module 2 has its air inlet connected to several sampling probes via pipelines with programmable sampling valves, for collecting, removing dust, and mixing the flue gas samples. The ammonia nitrogen and temperature integrated measurement module 3 is fixedly installed on the detection cell interface of the ammonia nitrogen and temperature integrated auxiliary module. It is used to measure the concentrations of NO, NO2, NH3, and O2 in the mixed flue gas, as well as the temperature, pressure, and flow rate, and generate real-time monitoring signals. The decoupling and coupling control module 4 includes several programmable sampling valves, several programmable backflush valves, and a control panel. The control panel is electrically connected to each programmable sampling valve and each programmable backflush valve to control the on / off state, so as to realize the uniform measurement coupling mode, the zoned rotation measurement decoupling mode, the alternating interlocking mode, and the backflush cleaning mode for flue gas. The ammonia injection automatic control module 5 is connected to the integrated ammonia-nitrogen-temperature measurement module and the unit's DCS system. It is used to output optimized ammonia injection quantity control commands and zoned ammonia injection adjustment suggestions based on the real-time monitoring signals and unit operating parameters.

[0025] The aforementioned full-section multi-point sampling module covers the flue gas cross-section with matrix-distributed sampling probes, effectively capturing the true situation of uneven NOx and NH3 concentration distribution in the flue gas. Each probe corresponds to a zone, providing spatial resolution for subsequent zone-based sampling and ammonia injection adjustment, and can identify local ammonia-nitrogen imbalance areas. It is particularly suitable for environments with large inlet / outlet flue gas dimensions and complex flow fields in SCR reactors, and the matrix layout ensures the sampling's inclusiveness across flow velocity and concentration gradients.

[0026] The integrated ammonia nitrogen and temperature auxiliary module combines flow convergence, dust removal, and mixing into one unit, simplifying external pretreatment equipment and reducing pipeline pressure loss and leakage risks. By mixing and homogenizing flue gas samples from multiple zones, it eliminates instantaneous fluctuations and local disturbances, providing the analyzer with stable and uniform gas. Combined with a programmable sampling valve, it allows independent control of the sampling on / off state for each channel, forming the basis for switching between modes such as equalization measurement, rotation measurement, and backflushing.

[0027] The integrated ammonia nitrogen and temperature measurement module can simultaneously obtain NO, NO2, NH3, and O2 concentrations, as well as temperature, pressure, and flow rate, eliminating the need for multiple analyzers and ensuring strong data consistency over time. Fixed at the auxiliary module's detection cell interface, it eliminates the need for long sample gas pipelines, resulting in fast response and minimal adsorption loss. The output real-time monitoring signal is directly used for ammonia injection calculation and zonal adjustment suggestions, serving as input to the control model.

[0028] The decoupling and coupling control module provides four operating modes: Average concentration coupling mode: Obtains the average concentration across the entire cross section for macroscopic adjustment of total ammonia injection volume; Zoned rotation measurement decoupling mode: Sequentially measure the concentration of each zone to diagnose ammonia nitrogen matching deviation and guide the adjustment of zone valves; Alternating Embedded Mode: Balances response speed and partition details, suitable for scenarios with rapidly changing loads; Backflush cleaning mode: The programmable backflush valve automatically cleans the probe and pipeline to prevent blockage and extend the maintenance cycle.

[0029] The aforementioned ammonia injection automatic control module receives unit operating parameters such as load, air volume, and coal quality, and combines them with real-time ammonia nitrogen concentration to perform feedforward and feedback composite control, adapting to wide load variations.

[0030] While ensuring denitrification efficiency, this system reduces ammonia consumption, minimizes ammonia slip, and mitigates the risk of ammonium bisulfate blockage in the air preheater. It not only provides total quantity commands but also outputs the correction direction (open wider / closer) for ammonia injection valves in each zone, achieving precise ammonia injection. This avoids excessive ammonia injection that could lead to material waste and corrosion of downstream equipment, while also reducing environmental risks caused by excessive emissions.

[0031] The above five modules form a complete closed loop of "sampling, preprocessing, measurement, mode switching and control execution", realizing refined denitrification control under wide load operation of the unit.

[0032] The full-section multi-point sampling module sets up a measurement section at both the inlet and outlet flues of the SCR reactor. Each measurement section has 3 or 4 measurement points in the width direction and 3 or 4 measurement points in the depth direction, forming an M×N gridded measurement point matrix, where M is 3 or 4 and N is 3 or 4. Furthermore, the ammonia nitrogen temperature integrated auxiliary module is installed at the intermediate elevation between the two independent reactor sides of the SCR reactor.

[0033] A 300MW unit's SCR inlet flue is 8m wide and 6m deep, using a 3×3 grid (9 sampling probes). At 80% load, the NH3 concentration measured at the upper left corner of the outlet grid was 9 ppm, while the concentration at the central area was only 1 ppm. Analysis of the corresponding inlet grid revealed that the NOx concentration at this zone's inlet was low, but the ammonia injection rate was not reduced in time. Maintenance personnel adjusted the opening of the ammonia injection branch pipe for this zone based on the grid data, reducing the NH3 concentration to 2.5 ppm and preventing localized ammonia escape exceeding the limit.

[0034] A 1000MW unit uses a 4×4 grid (16 probes). When the unit rapidly reduced its load from 100% to 40%, the inlet flow field became skewed. The NOx concentration in the original mainstream region decreased, but the NOx concentration in the lower right corner increased. The gridded sampling completely captured this shift. The control system automatically adjusted the total ammonia injection and provided correction suggestions for each zone. The NOx emissions remained stable below 35mg / m³ throughout the process, with no instantaneous exceedances.

[0035] The distance between two SCR reactors (side A and side B) in a power plant is approximately 6 meters, with a steel structure platform in between, its elevation level with the sampling probe lead-out point. After installing the integrated ammonia nitrogen and temperature auxiliary module on this platform, the sample gas pipeline length from the farthest sampling point on side A to the module is 9 meters, and the farthest point on side B is 8 meters. Compared to the original scheme (where the module was placed in an analysis cabin 50 meters away), the NH3 measurement value improved from being 3 ppm too low to having a deviation of less than 0.3 ppm, and the response time was shortened from 35 seconds to 7 seconds.

[0036] At one site, the auxiliary module was installed near the outer wall of the flue, where the summer ambient temperature reached as high as 75℃, causing the programmable sampling valve to fail an average of 2 times per month. After being moved to a mid-elevation position, where the ambient temperature dropped to around 45℃, it operated continuously for 6 months without any valve coil burnout. Simultaneously, maintenance personnel could operate both backflushing valves and sampling valves on the mid-platform, reducing the time for each backflushing and cleaning operation from 2 hours (two people per side) to 40 minutes per person.

[0037] The integrated ammonia nitrogen and temperature auxiliary module incorporates a high-temperature flue gas cyclone dust collector or gravity settling dust collector and is equipped with a compressed air preheating backflushing interface, which is connected to the backflushing valve in the decoupling and coupling control module. By integrating dust removal and preheating backflushing functions, it effectively prevents dust in the sample gas from depositing and scaling in the pipeline and detection pool, and uses preheated compressed air to remove accumulated dust, significantly reducing maintenance frequency and ensuring long-term measurement stability.

[0038] The ammonia injection automatic control module adopts a fanless, low-power industrial controller, supports the Modbus RTU communication protocol, and achieves redundant and seamless switching connection with the unit's DCS system. When the ammonia injection automatic control module fails, control of the ammonia injection regulating valve is automatically and seamlessly returned to the DCS native control system. This automatic and seamless transfer of control of the ammonia injection regulating valve to the unit's DCS system prevents ammonia injection interruption or sudden regulation changes due to a single module failure, ensuring the continuous and safe operation of the denitrification system.

[0039] The present invention also relates to a denitrification method based on the aforementioned denitrification equipment adapted to wide-load operation of the unit, comprising the following steps: S1. Multimodal homogeneous sampling and real-time monitoring: First, the sampling valves of all zones are opened to mix the flue gas across the entire cross-section and then perform integrated measurement to obtain the average parameters of the entire cross-section. Then, according to a preset time period, the sampling valves of other zones are closed sequentially, and only the sampling valves of a single zone are opened to perform rotational measurement on that zone to obtain the zone characteristic parameters. By combining the average measurement coupling with the rotational measurement of zones, both the average parameters of the entire cross-section are obtained for total quantity control, and the characteristic parameters of each zone are obtained for fine adjustment, realizing the unity of "global steady-state optimization" and "local dynamic diagnosis", providing a high-resolution data foundation for subsequent model prediction and zoned ammonia injection.

[0040] S2. Construct and run a wide-load adaptive prediction model: Using unit load, inlet NOx concentration, flue gas temperature, flue gas flow rate, historical ammonia injection rate, and historical ammonia slip concentration as input features, establish and train a deep neural network model to predict the outlet NOx concentration and ammonia slip concentration for the next 3-10 minutes; the deep neural network model performs online incremental updates based on actual measurement data during unit operation; by adopting a deep neural network and realizing online incremental updates, it can adapt to dynamic characteristics such as unit load changes, coal quality fluctuations, and equipment aging, accurately predicting outlet NOx and ammonia slip 3-10 minutes in advance, providing reliable input for feedforward control, and significantly reducing the lag effect of traditional feedback control.

[0041] S3. Perform multi-objective constraint optimization control: Using the environmental NOx emission limit as a hard constraint, and minimizing the ammonia injection rate and maximizing the denitrification efficiency as dual optimization objectives, the model prediction value from step S2 is used for feedforward control, combined with feedback PID control based on real-time tuning parameters using model gradient information, to jointly calculate the total ammonia injection rate command; simultaneously, based on the characteristic parameters of each zone obtained in step S1, adjustment suggestions for each branch of the ammonia injection grid are calculated and output; under the premise of meeting the environmental emission hard constraint, with minimizing the ammonia injection rate and maximizing the denitrification efficiency as dual objectives, combining feedforward prediction and gradient tuning feedback PID, and simultaneously outputting zone adjustment suggestions, the synergistic optimization of global economy and local ammonia nitrogen matching is achieved, effectively reducing ammonia consumption and ammonia escape risk.

[0042] S4. Wide-Load Adaptive Switching and Execution: Real-time monitoring of unit load change rate. When the load change rate exceeds a preset threshold, the control strategy is automatically switched to feedforward-dominant mode to reduce the weight of feedback control. When the load stabilizes, it switches back to normal control mode and executes the calculated ammonia injection command through the control system. The automatic switching of control strategy (feedforward-dominant mode and normal mode) based on the load change rate suppresses feedback overshoot and avoids instantaneous emission exceedances during rapid unit load changes, while resuming fine-tuning in steady state to ensure the response speed and stability of the denitrification system under all operating conditions.

[0043] The deep neural network model in step S2 is a BP neural network with two hidden layers; its input layer feature vector Specifically, this includes: real-time load of the generating unit. Average NOx concentration at SCR inlet Average temperature of flue gas inside the reactor flue gas flow rate Total ammonia injection volume at the previous moment The average ammonia escape concentration at the export point at the previous moment Output layer prediction vector The predicted NOx concentration at the outlet at time τ in the future. Predicted ammonia escape concentration The model was trained using the mean squared error loss function and the Adam optimizer, with an initial learning rate of 0.001 and 500 training epochs.

[0044] The specific integration of this deep neural network model (a 2-layer hidden-layer BP neural network) with the denitrification control scenario is as follows: During wide-load operation of the unit, the SCR reactor faces control challenges such as large lag (it typically takes tens of seconds to several minutes for flue gas to flow from the ammonia injection grid to the outlet NOx measuring point), strong nonlinearity (reaction efficiency varies drastically with temperature and load), and multivariate coupling (the ammonia injection rate simultaneously affects both outlet NOx and ammonia slip). The model utilizes measurable variables from the current and previous moments to predict the outlet NOx and ammonia slip concentrations 3-10 minutes in advance, providing a feedforward reference for ammonia injection control, thereby overcoming measurement lag and achieving "advanced adjustment." The online incremental update mechanism enables the model to adapt to long-term drift caused by catalyst aging, coal type changes, etc.

[0045] Input variables: Output variables: Scenario: A 600MW unit is operating stably at 85% load. The measured data at time t are as follows: Load L(t) = 510 MW; Inlet =380 mg / m³; flue gas temperature =385 ℃; flue gas flow rate ; Ammonia injection amount at the previous moment =95 kg / h; ammonia escape a moment ago =2.1 ppm.

[0046] Model (τ=5 minutes) predicted output: Export NOx forecast =32 mg / m³ (lower than the emission limit of 40 mg / m³); Ammonia escape prediction value =3.8 ppm (close to the warning value of 4 ppm).

[0047] Control Action: Due to the predicted impending ammonia slip exceeding the limit, the automatic ammonia injection control module, while maintaining NOx compliance, gradually reduced the ammonia injection rate from 95 kg / h to 88 kg / h through multi-objective optimization calculations. Five minutes later, the measured outlet NOx was 34 mg / m³, and the ammonia slip was 2.9 ppm, verifying the accuracy of the model's prediction. Without this prediction, relying solely on feedback control, reducing the ammonia injection rate only after the measured slip reached 4 ppm would have resulted in approximately 10 minutes of excessive ammonia injection, increasing the risk of air preheater blockage.

[0048] Training data: Collect historical operating data of the unit (covering 30%~100% load range, different coal types, and different ambient temperatures), and construct sample pairs according to time series: [X(t), Y(t+\tau)], where τ is 3~10 minutes (calibrated according to the actual lag time of flue gas from the ammonia injection grid to the outlet measuring point).

[0049] Loss function: Mean squared error (MSE), which penalizes both NOx and escape prediction bias.

[0050] Optimizer: Adam, initial learning rate 0.001, training for 500 epochs, early stopping method is used to prevent overfitting.

[0051] Online incremental update: Every hour of operation, newly collected measured data (actual outlet NOx and ammonia slip) are added to the training set to perform small-batch incremental training on the model (e.g., triggered once every 24 hours, or triggered when the prediction error continuously exceeds the threshold), so that the model can adapt to catalyst activity decay or seasonal coal quality changes.

[0052] In step S3, the total ammonia injection command is calculated. The formula is: Among them, feedforward quantity Kstoich represents the stoichiometric correction coefficient identified online by the adaptive prediction model, which is dimensionless; NO xin The average NOx concentration at the SCR reactor inlet is expressed in mg / Nm³; FlueGas represents the flue gas flow rate in Nm³ / h; MNO x MNH3 represents the molar mass of nitrogen oxides; SPNO represents the molar mass of ammonia. x This represents the dynamically optimized NOx target value at the outlet, expressed in mg / Nm³. The feedback correction ammonia injection quantity The calculation formula is: ;in, , , These represent the proportional coefficient, integral coefficient, and differential coefficient, respectively, which change with time t. The sensitivity gradient of the ammonia injection rate is based on the output of the adaptive prediction model. Perform real-time adjustment; This represents the deviation of the outlet NOx concentration at time t, expressed in mg / Nm³, and is calculated using the following formula: ; This represents the measured NOx concentration at the SCR reactor outlet at time t, in mg / Nm³.

[0053] In the SCR denitrification process, there is a large lag (typically 30-120 seconds) and strong nonlinearity (reaction efficiency varies drastically with temperature, load, and catalyst activity) between the ammonia injection and the outlet NOx response. Traditional fixed-parameter PID controllers are difficult to adapt to wide-load operation and are prone to overshoot or oscillation.

[0054] This algorithm employs an adaptive feedforward strategy combined with model gradient tuning feedback. Feedforward quantity QFF: Based on inlet NOx, flue gas flow rate and target emission value, the theoretical ammonia injection quantity is calculated through the stoichiometric ratio correction factor Kstoich, so as to quickly track load and inlet concentration changes.

[0055] Feedback quantity QFB: It adopts a PID structure, but its proportional, integral, and derivative coefficients are not fixed. Instead, they are based on the sensitivity gradient of the ammonia injection quantity to the output of the adaptive prediction model (i.e., the deep neural network in step S2). Real-time tuning is performed. When model predictions indicate that changes in ammonia injection rate have a significant impact on outlet NOx, the PID gain is increased; when the impact is small, the gain is decreased, thereby adaptively adjusting the control strength.

[0056] This algorithm effectively solves the parameter tuning problem of traditional PID in objects with large time delay and nonlinearity, and is closely coupled with the prediction model in step S2 to form a "prediction-gradient-control" closed loop.

[0057] Feedforward Fast response to ingress disturbances; feedback quantity Eliminate steady-state errors and unmodeled dynamics. Feedback PID parameters. Instead of relying on experience to set parameters, the parameters change dynamically with the model gradient, thus achieving a match between control gain and object sensitivity.

[0058] Model gradient By directly linking the predictive model with the control algorithm, the control strategy becomes "predictive": if the model predicts that the future outlet NOx is very sensitive to changes in ammonia injection, the controller will automatically reduce the action to avoid over-adjustment.

[0059] Scenario: A 600MW unit is rapidly reducing its load from 70% to 50%. The inlet NOx concentration drops from 350 mg / m³ to 220 mg / m³, and the flue gas temperature drops from 380℃ to 340℃ (close to the lower limit of the reaction window).

[0060] At time t=0: measured outlet NOx = 38 mg / m³, target value SP = 35 mg / m³, deviation e = -3 mg / m³ (actual value is higher than target).

[0061] The deep neural network model provides the current gradient. This indicates that for every 1 kg / h increase in ammonia injection, the NOx output decreases by approximately 0.15 mg / m³ (moderate sensitivity).

[0062] Algorithm actions: 1. Feedforward calculation: .in The online model identification value is 0.92 (due to the low temperature, efficiency decreases, and the theoretical ammonia quantity needs to be increased). The feedforward provides a basic ammonia injection rate of 85 kg / h.

[0063] 2. Gradient tuning: Based on the sensitivity of -0.15, the controller automatically sets the proportional coefficient. (Medium gain), integral coefficient Differential coefficients .

[0064] 3. Feedback Calculation: With deviation e = -3, the PID controller calculates the feedback correction amount. = -2.1kg / h (reduced ammonia injection).

[0065] 4. General Instructions: The order was issued and implemented.

[0066] Two minutes later: the model gradient decreased to -0.05 due to the further drop in temperature (reduced sensitivity). The controller automatically increased K_p to 2.5 to enhance the feedback effect and prevent the outlet NOx from exceeding the limit due to the decrease in reaction efficiency. Finally, the measured outlet NOx stabilized at 34~36 mg / m³, and ammonia slip did not exceed the limit.

[0067] Compared to traditional fixed PID: If With a fixed value of 1.5, the feedback effect is too weak when the sensitivity drops to -0.05, and the outlet NOx may rise to 42 mg / m³, exceeding the standard. If the value is fixed at 2.5, ammonia injection oscillation is easily caused when the sensitivity is -0.15. This algorithm achieves stable control under a wide load by adaptively matching different operating conditions through real-time gradient tuning.

[0068] Step S3 further includes: when the ammonia escape prediction value output by the adaptive prediction model... Or measured ammonia slip value When the ammonia escape rate exceeds 3 ppm, the protection function is triggered, blocking any further increases in the ammonia injection regulating valve and forcibly reducing the total ammonia injection rate by 5% until the ammonia escape value drops below 2.5 ppm. By automatically locking and forcibly reducing the ammonia escape rate when it exceeds the limit, the system can effectively prevent the air preheater from becoming clogged with ammonium bisulfate and downstream equipment from corrosion caused by excessive ammonia injection, thus ensuring the long-term safe operation of the denitrification system.

[0069] In step S1, the average measurement mode and the rotation measurement mode are executed in an interleaved manner. The duration of the average measurement mode is 50 to 70 minutes, and the duration of the rotation measurement mode for each single zone is 8 to 12 minutes. Furthermore, the backflushing cleaning mode is triggered by a timed trigger or by a backflushing request signal from the integrated measurement module. During backflushing, all programmable sampling valves are closed, and programmable backflushing valves are opened, performing pulse-type cleaning of the sampling pipeline and the integrated auxiliary module. Simultaneously, the measurement module maintains the previous value output. By interleaving long-cycle average measurement (50-70 minutes) with short-cycle zone rotation measurement (8-12 minutes per zone), the continuous stability of the average parameters across the entire cross-section is ensured, while the ammonia nitrogen deviation in each zone can be quickly captured. At the same time, maintaining the previous value output during backflushing avoids measurement value jumps or control disturbances caused by the cleaning action.

[0070] Long-cycle (50-70 minutes) average measurement mode: Covers the normal load fluctuation cycle of the unit, provides a stable full-section average NOx and ammonia slip benchmark, and avoids control jitter caused by frequent mode switching.

[0071] Single-zone round-robin short cycle (8~12 minutes): It can complete a round scan of all zones in a short time (e.g., 9 zones in a 3×3 grid, 10 minutes each, a round is about 90 minutes), and promptly detect local uneven ammonia injection or catalyst blockage, meeting the timeliness requirements of zone adjustment in engineering.

[0072] Maintaining the previous output value during backflushing: When backflushing, the sample gas path is cut off. If an invalid value or a value is output directly to zero, it will cause drastic fluctuations in ammonia injection control. Maintaining the previous valid value can ensure the continuity of control commands. Normal measurement can be quickly resumed after backflushing is completed, achieving uninterrupted dust removal.

[0073] The preset threshold in step S4 is set to an absolute value of the load change rate greater than 3% / minute; in the feedforward-dominated mode, the feedforward quantity... In general instructions The calculation weight in the process is increased to 80% to 100%, and the integral term of the feedback PID controller is frozen. When the unit load change rate exceeds 3% / minute, by increasing the feedforward weight to 80%~100% and freezing the feedback integral term, the overshoot and integral saturation of the feedback control under large disturbances can be effectively suppressed, enabling the ammonia injection quantity to quickly follow the load change and avoiding instantaneous NOx exceedance at the outlet.

[0074] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A denitrification device adapted to wide-load operation of a generating unit, characterized in that, include: A full-section multi-point sampling module includes several sampling probes arranged in a matrix on the cross-section of the SCR reactor inlet flue and outlet flue, which are used to collect flue gas samples from different zones. An integrated ammonia nitrogen and temperature auxiliary module has its air inlet connected to several sampling probes via pipelines equipped with programmable sampling valves, for collecting, removing dust, and mixing the flue gas samples. An integrated ammonia nitrogen and temperature measurement module is fixedly installed on the detection cell interface of the integrated ammonia nitrogen and temperature auxiliary module. It is used to measure the concentrations of NO, NO2, NH3, and O2 in the mixed flue gas, as well as the temperature, pressure, and flow rate, and to generate real-time monitoring signals. The decoupling and coupling control module includes several programmable sampling valves, several programmable backflush valves, and a control panel. The control panel is electrically connected to each programmable sampling valve and each programmable backflush valve to control the on / off state, so as to realize the uniform measurement coupling mode, the zoned rotation measurement decoupling mode, the alternating interlocking mode, and the backflush cleaning mode for flue gas. The ammonia injection automatic control module is connected to the integrated ammonia-nitrogen-temperature measurement module and the unit's DCS system. It is used to output optimized ammonia injection quantity control commands and zoned ammonia injection adjustment suggestions based on the real-time monitoring signals and unit operating parameters.

2. The denitrification equipment adapted to wide-load operation of the unit according to claim 1, characterized in that, The full-section multi-point sampling module sets up a measurement section at both the inlet and outlet flues of the SCR reactor. Each measurement section has 3 or 4 measurement points in the width direction and 3 or 4 measurement points in the depth direction, forming an M×N gridded measurement point matrix, where M is 3 or 4 and N is 3 or 4. Furthermore, the ammonia nitrogen temperature integrated auxiliary module is installed at the intermediate elevation between the two independent reactor sides of the SCR reactor.

3. The denitrification equipment adapted to wide-load operation of the unit according to claim 1, characterized in that, The integrated ammonia nitrogen and temperature auxiliary module integrates a high-temperature flue gas cyclone dust removal or gravity settling dust removal component, and is equipped with a compressed air preheating backflush interface, which is connected to the backflush valve in the decoupling and coupling control module.

4. The denitrification equipment adapted to wide-load operation of the unit according to claim 1, characterized in that, The ammonia injection automatic control module adopts a fanless, low-power industrial controller, supports the Modbus RTU communication protocol, and achieves redundant and seamless switching connection with the unit's DCS system. When the ammonia injection automatic control module fails, the control of the ammonia injection regulating valve is automatically and seamlessly returned to the DCS native control system.

5. A denitrification method based on the denitrification equipment adapted to wide-load operation of a unit according to any one of claims 1 to 4, characterized in that, Includes the following steps: S1. Multimodal homogeneous sampling and real-time monitoring: First, open the sampling valves of all zones, mix the flue gas of the entire cross section and perform integrated measurement to obtain the average parameters of the entire cross section; then, according to the preset time period, close the sampling valves of other zones in sequence and open only a single zone to perform round-robin measurement on the zone to obtain the zone characteristic parameters. S2. Construct and run a wide-load adaptive prediction model: Using unit load, inlet NOx concentration, flue gas temperature, flue gas flow rate, historical ammonia injection rate, and historical ammonia slip concentration as input features, establish and train a deep neural network model to predict the outlet NOx concentration and ammonia slip concentration for the next 3-10 minutes; the deep neural network model is updated online incrementally based on actual measurement data during unit operation; S3. Perform multi-objective constraint optimization control: With the environmental NOx emission limit as a hard constraint, and minimizing the ammonia injection amount and maximizing the denitrification efficiency as dual optimization objectives, the model prediction value from step S2 is used for feedforward control, and combined with feedback PID control based on model gradient information to tune parameters in real time, the total ammonia injection amount command is calculated together; at the same time, based on the characteristic parameters of each zone obtained in step S1, the adjustment suggestions for each branch of the ammonia injection grid are calculated and output. S4. Wide-load condition adaptive switching and execution: Real-time monitoring of unit load change rate. When the load change rate exceeds the preset threshold, the control strategy is automatically switched to feedforward dominant mode to reduce the weight of feedback control. Once the load stabilizes, switch back to normal control mode and issue the calculated ammonia injection command through the control system for execution.

6. The denitrification method according to claim 5, characterized in that, The deep neural network model in step S2 is a BP neural network with two hidden layers; its input layer feature vector Specifically, this includes: real-time load of the generating unit. Average NOx concentration at SCR inlet Average temperature of flue gas inside the reactor flue gas flow rate Total ammonia injection volume at the previous moment The average ammonia escape concentration at the export point at the previous moment Output layer prediction vector The predicted NOx concentration at the outlet at time τ in the future. Predicted ammonia escape concentration The model was trained using the mean squared error loss function and the Adam optimizer, with an initial learning rate of 0.001 and 500 training epochs.

7. The denitrification method according to claim 5, characterized in that, In step S3, the total ammonia injection command is calculated. The formula is: Among them, feedforward quantity Kstoich represents the stoichiometric correction coefficient identified online by the adaptive prediction model, which is dimensionless; NO xin The average NOx concentration at the SCR reactor inlet is expressed in mg / Nm³; FlueGas represents the flue gas flow rate in Nm³ / h; MNO x MNH3 represents the molar mass of nitrogen oxides; SPNO represents the molar mass of ammonia. x This represents the dynamically optimized outlet NOx target value, in mg / Nm³. The feedback correction ammonia injection quantity The calculation formula is: ;in, , , These represent the proportional coefficient, integral coefficient, and differential coefficient, respectively, which change with time t. The sensitivity gradient of the ammonia injection rate is based on the output of the adaptive prediction model. Perform real-time adjustment; This represents the deviation of the outlet NOx concentration at time t, expressed in mg / Nm³, and is calculated using the following formula: ; This represents the measured NOx concentration at the SCR reactor outlet at time t, in mg / Nm³.

8. The denitrification method according to claim 5, characterized in that, Step S3 further includes: when the ammonia escape prediction value output by the adaptive prediction model... Or measured ammonia slip value When the ammonia escape rate exceeds 3 ppm, the protection function is triggered, blocking any further increase in the ammonia injection regulating valve and forcibly reducing the total ammonia injection rate by 5% until the ammonia escape value drops below 2.5 ppm.

9. The denitrification method according to claim 5, characterized in that, The uniform measurement mode and the rotation measurement mode in step S1 are executed in an interleaved manner. The duration of the uniform measurement mode is 50 to 70 minutes, and the duration of the rotation measurement mode for each single zone is 8 to 12 minutes. Furthermore, the backflushing cleaning mode is triggered by a timer or by the backflushing request signal from the integrated measurement module. During the backflushing, all programmable sampling valves are closed, and the programmable backflushing valves are opened to perform pulse cleaning on the sampling pipeline and the integrated auxiliary module. At the same time, the measurement module maintains the previous value output.

10. The denitrification method according to claim 5, characterized in that, The preset threshold in step S4 is set to an absolute value of the load change rate greater than 3% / minute; in the feedforward-dominated mode, the feedforward quantity... In general instructions The computational weight in the process is increased to 80% to 100%, and the integral term of the feedback PID controller is frozen.

Images