A method and system for feedforward and feedback combined control of ammonia injection quantity in an SCR denitrification system

Abstract

This invention relates to the field of SCR denitrification control, and discloses a method and system for feedforward feedback composite control of ammonia injection rate in an SCR denitrification system. This method constructs a catalyst temperature storage index tree by collecting historical and real-time operating data of the SCR system to form a sample set of operating conditions. Based on the index tree and the real-time temperature change rate, it infers the ammonia adsorption and desorption flux distribution to obtain a dynamic response diagram of ammonia storage. It calculates the equivalent stored ammonia flow rate and superimposes it with the theoretical ammonia injection rate to generate an ammonia injection feedforward compensation sequence, which is then input into the controller to generate ammonia injection control commands and drive the actuators. Simultaneously, it updates the index tree online based on the prediction deviation. This invention, by parameterizing the catalyst ammonia storage capacity and proactively compensating for the adsorption and desorption flux caused by temperature changes, solves the problem of excessive or insufficient ammonia caused by rapid temperature changes, achieving stable control of outlet nitrogen oxides and ammonia escape during variable load and temperature processes.

Description

Technical Field

[0001] This invention relates to the field of SCR denitrification control, and more particularly to a method and system for feedforward and feedback composite control of ammonia injection quantity in an SCR denitrification system. Background Technology

[0002] During the variable load operation of thermal power units, a feedforward feedback control system is typically used to control nitrogen oxide emissions. This system tracks changes in inlet and outlet nitrogen oxide concentrations and flue gas flow rates, adjusts the ammonia injection rate to achieve denitrification, and compensates for transmission delays to maintain the dynamic balance of the denitrification process.

[0003] In actual operation, when the unit undergoes large-scale load changes and flue gas temperature changes rapidly, especially during the rapid rise in flue gas temperature, although the ammonia injection rate output by the controller does not reach the theoretical excess value, the actual ammonia slip rate suddenly increases. Conversely, during the rapid drop in flue gas temperature, the ammonia injection rate, according to theoretical calculations, meets the requirements, but the outlet nitrogen oxide concentration temporarily exceeds the standard and cannot be suppressed in time. The root cause of this phenomenon is that the existing control model uses static calculation logic based on the stoichiometric ratio of transient chemical reactions, failing to consider the dynamic adsorption and storage effect of ammonia on the SCR catalyst as a porous medium. The saturated ammonia coverage on the catalyst surface changes with temperature; when the temperature rises, the stored ammonia desorbs and releases, and when the temperature falls, it adsorbs the injected ammonia, causing a deviation between the actual amount of ammonia participating in the reaction and the controller's calculated value. This deviation can cause a sudden increase in ammonia slip rate, potentially leading to blockage of downstream equipment, and an excess of outlet nitrogen oxide concentration, potentially causing environmental emission problems. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A method for controlling ammonia injection in a SCR system based on a catalyst temperature storage index tree includes:

[0006] S1, collect historical and real-time operating data of the SCR system and form a set of operating condition feature samples, and build a catalyst temperature storage index tree based on the set of operating condition feature samples;

[0007] S2, based on the catalyst temperature storage index tree and real-time temperature change rate, the ammonia adsorption and desorption flux distribution is inferred to obtain the ammonia storage dynamic response map;

[0008] S3. Calculate the equivalent stored ammonia flow rate based on the ammonia storage dynamic response diagram and superimpose it with the theoretical ammonia injection rate to generate a distribution-type ammonia injection feedforward compensation sequence.

[0009] S4. Input the distributed ammonia injection feedforward compensation sequence into the SCR feedforward feedback composite controller, generate the final ammonia injection control command based on the distributed ammonia injection feedforward compensation sequence, and drive the ammonia injection actuator to complete online adjustment.

[0010] Furthermore, S1 includes the following steps:

[0011] S11. Based on the signal point list of the SCR system, the acquisition channels of the distributed control system, the acquisition channels of the flue gas online monitoring system, and the feedback channels of the ammonia injection actuator are uniformly mapped to obtain a multi-source signal mapping table.

[0012] S12, based on the multi-source signal mapping table, performs synchronous processing and completion on historical and real-time acquired data to obtain a complete running sample matrix;

[0013] S13, perform outlier removal processing based on the complete running sample matrix to obtain the cleaned running sample matrix;

[0014] S14, Perform working condition segmentation processing based on the cleaning operation sample matrix to obtain the working condition feature sample set;

[0015] S15, based on the working condition feature sample set and the steady-state working condition segment set obtained in the working condition segmentation process, perform steady-state ammonia storage parameter identification processing to obtain the steady-state ammonia storage parameter table.

[0016] S16, Based on the working condition feature sample set and the rapid temperature change working condition segment set obtained in the working condition segmentation process, perform dynamic parameter identification processing for temperature change ammonia storage to obtain the dynamic parameter table for temperature change ammonia storage.

[0017] S17. Based on the steady-state ammonia storage parameter table and the variable-temperature ammonia storage dynamic parameter table, a catalyst temperature storage index tree is constructed to obtain the catalyst temperature storage index tree.

[0018] Furthermore, S12 includes the following steps:

[0019] S121, Based on the multi-source signal mapping table, perform synchronous acquisition and archiving processing on historical acquisition data and real-time acquisition data to obtain a set of multi-source raw operating data;

[0020] S122, based on the data acquisition timestamps in the multi-source original running data set, the sampling sequences of each variable are resampled to obtain a unified time axis running sample matrix;

[0021] S123, perform missing point imputation processing on the running sample matrix based on the unified time axis to obtain the complete running sample matrix;

[0022] The resampling process employs piecewise linear interpolation with added monotonicity constraints. These constraints limit the increment of the interpolation point to the supremum of the increment of the adjacent original sampling point. The supremum is calculated from the upper quantile of the distribution of the absolute values ​​of the increments of the corresponding variables in the multi-source original running data set.

[0023] Furthermore, S14 includes the following steps:

[0024] S141, calculate the temperature change rate matrix and flow rate change rate sequence based on the cleaning operation sample matrix to obtain the change rate feature matrix;

[0025] S142, calculate the steady-state discrimination threshold and the rapid temperature change discrimination threshold based on the rate of change feature matrix to obtain the segmented threshold parameters;

[0026] S143, Based on the segmented threshold parameter, the cleaning operation sample matrix is ​​divided into segments to obtain a set of steady-state operating condition segments and a set of rapid temperature change operating condition segments.

[0027] S144, Based on the set of steady-state operating condition segments and the set of rapid temperature change operating condition segments, sample annotation processing is performed on the cleaning operation sample matrix to obtain the operating condition feature sample set;

[0028] The set of steady-state operating condition segments is a set of time intervals in which both the rate of temperature change and the rate of flow change do not exceed the steady-state discrimination threshold, and the set of rapid temperature change operating condition segments is a set of time intervals in which the rate of temperature change exceeds the rapid temperature change discrimination threshold and the duration exceeds the minimum segment duration.

[0029] Furthermore, S17 includes the following steps:

[0030] S171, Based on the structural parameters of the SCR reactor and the catalyst layer arrangement parameters, the catalyst spatial unit set is divided to obtain the catalyst spatial unit index table;

[0031] S172, Based on the catalyst spatial unit index table, perform spatial binding processing on the steady-state ammonia storage parameter table and the variable-temperature ammonia storage dynamic parameter table to obtain a spatially bound ammonia storage parameter set;

[0032] S173, based on the spatially bound ammonia storage parameter set, calculate the confidence parameter and deactivation coefficient to obtain the confidence deactivation parameter set;

[0033] S174, Based on the spatially bound ammonia storage parameter set and the confidence deactivation parameter set, a tree node record is constructed to obtain the catalyst temperature storage index tree node set;

[0034] The catalyst temperature storage index tree is a hierarchical tree structure. The root node corresponds to the entire SCR reactor, the first-level nodes correspond to the catalyst layer number, the second-level nodes correspond to the temperature range, the third-level nodes correspond to the flue gas space velocity range, and the leaf nodes store tree node records. Each tree node record includes at least the ammonia saturation storage capacity parameter, the equilibrium ammonia coverage parameter, the storage capacity temperature slope parameter, the dynamic response time parameter, the reliability parameter, and the deactivation coefficient, and uses the code in the catalyst spatial unit index table as the index key.

[0035] Furthermore, S2 includes the following steps:

[0036] S21, based on the unified time axis running sample matrix, read the running sample vector at the current moment from the real-time acquisition window to obtain the real-time running sample vector;

[0037] S22, perform differential smoothing processing on the historical samples corresponding to the preset number of review steps in the real-time running sample vector and the unified time axis running sample matrix to obtain the real-time temperature change rate vector and the real-time flow change rate value.

[0038] S23, short-time temperature prediction processing is performed based on the real-time temperature change rate vector and the real-time running sample vector to obtain a short-time temperature prediction sequence;

[0039] S24, calculate the flue gas space velocity prediction sequence based on the short-time temperature prediction sequence and the real-time running sample vector to obtain the space velocity prediction sequence;

[0040] S25, based on the short-time temperature prediction sequence, space velocity prediction sequence and catalyst spatial unit index table, perform index retrieval processing to obtain the local ammonia storage parameter vector set;

[0041] S26, perform equilibrium storage update processing based on the local ammonia storage parameter vector set and short-time temperature prediction sequence to obtain the target equilibrium ammonia storage sequence;

[0042] S27. Based on the target balanced ammonia storage sequence and the balanced ammonia coverage parameter, perform initial ammonia coverage state quantity estimation processing to obtain the initial ammonia coverage state quantity set.

[0043] S28. Based on the initial set of ammonia coverage state variables, the target balanced ammonia storage sequence and the dynamic response time parameter, perform dynamic convergence calculation to obtain the ammonia coverage state variable prediction sequence.

[0044] S29, Based on the ammonia coverage state quantity prediction sequence and the target equilibrium ammonia storage quantity sequence, perform storage ammonia change calculation processing to obtain a set of local storage ammonia change quantity prediction sequences;

[0045] S210, based on the locally stored ammonia change prediction sequence set and the sampling period, flux conversion processing is performed to obtain the ammonia adsorption and desorption flux distribution;

[0046] S211, based on the ammonia adsorption and desorption flux distribution and the catalyst spatial unit index table, a transport delay mapping process is performed to obtain the unit transport delay table;

[0047] S212, based on the ammonia adsorption and desorption flux distribution, the ammonia coverage state quantity prediction sequence and the unit transport delay table, a dynamic response diagram of ammonia storage is constructed, and the dynamic response diagram of ammonia storage is obtained.

[0048] Furthermore, the ammonia storage dynamic response diagram is a directed graph data structure. The nodes of the ammonia storage dynamic response diagram are jointly encoded by the catalyst spatial unit index key and the prediction time index. The node attributes include at least the predicted temperature value, the predicted space velocity value, the predicted ammonia coverage state quantity value, and the ammonia adsorption and desorption flux value. The directed edge of the ammonia storage dynamic response diagram connects the upstream nodes and the downstream nodes along the flue gas flow direction, and the flux influence of the upstream node is mapped to the corresponding prediction time index of the downstream node according to the unit transmission delay table.

[0049] Furthermore, S3 includes the following steps:

[0050] S31, based on the target values ​​of inlet nitrogen oxide concentration and outlet nitrogen oxide concentration in the real-time running sample vector and the current inlet flue gas flow rate of the SCR reactor, perform stoichiometric conversion to obtain the theoretical ammonia injection rate;

[0051] S32, based on the outlet nitrogen oxide concentration and ammonia slip concentration in the real-time running sample vector, performs feedback control calculation and processing to obtain the feedback correction ammonia injection amount;

[0052] S33, based on traversing the directed edges of the rolling ammonia storage dynamic response graph and accumulating the flux contribution, the equivalent storage ammonia flow sequence is obtained;

[0053] S34, based on the equivalent stored ammonia flow rate sequence and the theoretical ammonia injection rate, a feedforward compensation synthesis process is performed to obtain the ammonia injection feedforward compensation sequence;

[0054] S35, based on the real-time temperature change rate vector and the unit load change rate, calculate the feedforward compensation weight coefficient and the feedback correction weight coefficient to obtain the set of weight coefficients;

[0055] S36, based on the ammonia injection feedforward compensation sequence, the feedback correction ammonia injection quantity and the weight coefficient set, perform composite setpoint fusion processing to obtain the composite ammonia injection setpoint sequence;

[0056] S37, based on the composite ammonia injection setpoint sequence and the layered ammonia injection flow constraint table, perform allocation calculation processing to obtain the allocation type ammonia injection feedforward compensation sequence;

[0057] The equivalent storage ammonia flow sequence is obtained by mapping and accumulating the ammonia adsorption and desorption flux values ​​on the directed edge pointing to the outlet direction in the ammonia storage dynamic response diagram according to the unit transmission delay table. The components greater than 0 in the equivalent storage ammonia flow sequence indicate that there is an equivalent ammonia supply from the catalyst storage in the future period, and the components less than 0 indicate that there is an equivalent ammonia deficiency caused by catalyst adsorption in the future period.

[0058] Furthermore, S4 includes the following steps:

[0059] S41, construct an ammonia injection feedforward compensation queue based on the allocation-type ammonia injection feedforward compensation sequence;

[0060] S42, based on the ammonia injection feedforward compensation queue, extract the first element of the current queue and perform benchmark superposition processing to obtain the current feedforward ammonia injection setpoint vector;

[0061] S43, based on the current feedforward ammonia injection setpoint vector and the feedback correction ammonia injection quantity, perform the final fusion processing to obtain the final ammonia injection control command;

[0062] S44, based on the final ammonia injection control command, executes the ammonia injection actuator drive processing to obtain the actuator drive signal;

[0063] S45, based on the actuator drive signal and the real-time running sample vector, perform rolling update processing on the ammonia injection feedforward compensation queue to obtain the updated ammonia injection feedforward compensation queue.

[0064] S46. Based on the rolling ammonia storage dynamic response diagram, the predicted outlet nitrogen oxide concentration and the predicted ammonia escape concentration are calculated to obtain the predicted emission sequence.

[0065] S47, based on the predicted emission sequence and the measured outlet nitrogen oxide concentration and measured ammonia slip concentration in the real-time operating sample vector, the deviation calculation process is performed to obtain the predicted deviation sequence;

[0066] S48, calculate the prediction deviation threshold based on the prediction deviation sequence and perform over-limit discrimination processing to obtain the online adaptive write-back request;

[0067] S49, based on the online adaptive write-back request and index tree update input queue, perform incremental write-back processing to obtain the updated catalyst temperature storage index tree;

[0068] S410, based on the updated catalyst temperature storage index tree and the real-time running sample vector, triggers the reconstruction of the rolling ammonia storage dynamic response map for the next control cycle, and obtains the rolling ammonia storage dynamic response map for the next cycle.

[0069] A feedforward feedback composite control system for ammonia injection quantity in an SCR denitrification system is provided to implement the aforementioned feedforward feedback composite control method for ammonia injection quantity in an SCR denitrification system. The system includes:

[0070] Data acquisition and index tree construction module: used to collect historical and real-time operating data of SCR system and form a set of operating condition feature samples, and build a catalyst temperature storage index tree based on the set of operating condition feature samples;

[0071] Ammonia storage dynamic response map generation module: used to infer the ammonia adsorption and desorption flux distribution based on the catalyst temperature storage index tree and real-time temperature change rate, and obtain the ammonia storage dynamic response map;

[0072] Ammonia injection feedforward compensation sequence generation module: used to calculate the equivalent stored ammonia flow rate based on the ammonia storage dynamic response diagram and superimpose it with the theoretical ammonia injection rate to generate a distribution type ammonia injection feedforward compensation sequence;

[0073] Ammonia injection control and execution module: used to input the distributed ammonia injection feedforward compensation sequence into the SCR feedforward feedback composite controller, generate the final ammonia injection control command based on the distributed ammonia injection feedforward compensation sequence, and drive the ammonia injection actuator to complete online adjustment.

[0074] This invention effectively solves the systemic deviation problem of ammonia escape surge and short-term nitrogen oxide exceedance at the outlet caused by the dynamic adsorption and desorption of catalyst ammonia storage during rapid temperature changes in SCR systems. This is achieved by constructing a catalyst temperature storage index tree and a rolling ammonia storage dynamic response map. The method unifies the encoding of steady-state ammonia storage parameters (equilibrium ammonia coverage, ammonia saturated storage capacity) and dynamic parameters (storage capacity temperature slope, dynamic response time) into a hierarchical index tree, and combines reliability parameters and deactivation coefficients to achieve reliable parameter retrieval. The dynamic response map explicitly models the local flux driven by the rate of temperature change and the along-path transport delay, forming a quantifiable equivalent stored ammonia flow sequence. Based on this, this invention achieves adaptive fusion of ammonia injection feedforward compensation and feedback control, reducing the risk of excessive ammonia desorption during the temperature rise stage and replenishing the ammonia demand during the temperature drop stage, thus suppressing the regulation oscillations and overshoot caused by measurement point lag in traditional control. Meanwhile, the online adaptive write-back mechanism dynamically updates the index tree parameters by predicting the deviation sequence, enabling the model to continuously correct itself as the catalyst deactivates and operating conditions change, ensuring the adaptability of the control strategy to the automatic power generation control peak shaving, load shifting, and temperature transition processes. The system's functional units cover data acquisition, parameter identification, dynamic response modeling, and control command generation, realizing look-ahead compensation and long-term stability maintenance of stored dynamic disturbances in SCR control. Attached Figure Description

[0075] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0076] Figure 1 This is a flowchart of a feedforward feedback composite control method for ammonia injection quantity in an SCR denitrification system according to the present invention;

[0077] Figure 2 This is a schematic diagram of data resampling and time axis alignment in an embodiment of the present invention;

[0078] Figure 3This is a schematic diagram illustrating the segmented processing of operating conditions and the steady-state rapid temperature change discrimination in an embodiment of the present invention;

[0079] Figure 4 This is a schematic diagram of the catalyst temperature storage index tree in an embodiment of the present invention;

[0080] Figure 5 This is a schematic diagram of the dynamic response of ammonia storage in an embodiment of the present invention;

[0081] Figure 6 This is a schematic diagram of the ammonia injection actuator drive in an embodiment of the present invention;

[0082] Figure 7 This is a schematic diagram of the ammonia injection feedforward compensation queue in an embodiment of the present invention;

[0083] Figure 8 This is a functional block diagram of a feedforward feedback composite control system for ammonia injection quantity in an SCR denitrification system according to the present invention. Detailed Implementation

[0084] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0085] Example 1:

[0086] Please see Figure 1 As shown, this embodiment provides a feedforward feedback composite control method for ammonia injection quantity in an SCR denitrification system, including:

[0087] S1: Collect historical and real-time operating data of the SCR system and form a set of operating condition characteristic samples. Construct a catalyst temperature storage index tree based on the set of operating condition characteristic samples.

[0088] Further, S1 includes the following steps:

[0089] S11: Based on the signal point list of the SCR system, the acquisition channels of the distributed control system, the acquisition channels of the flue gas online monitoring system, and the feedback channels of the ammonia injection actuator are uniformly mapped to obtain a multi-source signal mapping table. The multi-source signal mapping table is a key-value table with the signal name as the primary key and the sampling source address and engineering quantity unit as the value. The multi-source signal mapping table is used to merge signals that refer to the same physical quantity in different acquisition systems into the same variable name to ensure that the subsequent constructed working condition feature sample set meets the processing constraint of variable name consistency.

[0090] S12: Based on the multi-source signal mapping table, perform synchronous processing and completion on historical and real-time acquired data to obtain a complete running sample matrix;

[0091] Furthermore, the synchronization processing and completion includes the following steps:

[0092] S121: Based on the multi-source signal mapping table, historical and real-time acquired data are synchronously acquired and archived to obtain a multi-source original operating data set. The multi-source original operating data set is a set of records arranged in ascending order by timestamp. Each record includes at least the unit load, flue gas temperature at the inlet of each catalyst layer, flue gas temperature at the outlet of each catalyst layer, flue gas flow rate at the inlet of the SCR reactor, nitrogen oxide concentration at the inlet of the SCR reactor, nitrogen oxide concentration at the outlet of the SCR reactor, total ammonia injection flow rate, stratified ammonia injection flow rate, ammonia injection valve opening, ammonia escape concentration, and data acquisition timestamp. Synchronous acquisition ensures that physical quantities of the same source from different systems are archived with unified variable names. The archiving process retains the timestamp correlation and consistency of engineering quantity units of the original data.

[0093] S122: Based on the data acquisition timestamps in the multi-source original operational data set, the sampling sequences of each variable are resampled to obtain a unified time-axis operational sample matrix. This unified time-axis operational sample matrix uses the unified time axis as the row index and the variable name as the column index. The unified time axis is discretely generated by the sampling period set by the control system. The resampling process employs piecewise linear interpolation with added monotonicity constraints. These constraints limit the increment of the interpolation point to the supremum of the increment of adjacent original sampling points. The supremum is calculated from the upper quantile of the absolute value distribution of the corresponding variable's increment in the multi-source original operational data set, thereby suppressing non-physical oscillations introduced by interpolation. (See also...) Figure 2This is a schematic diagram of data resampling and time axis alignment provided in this application embodiment. As shown in the figure, the upper part shows the non-uniform distribution of the original multi-source data on the time axis. Circular markers represent signal points collected by the distributed control system, triangular markers represent signal points collected by the flue gas online monitoring system, and diamond markers represent signal points fed back by the ammonia injection actuator. These data points exhibit a clear characteristic of inconsistent sampling intervals on the time axis. The middle part of the figure illustrates the resampling technique using vertical dashed lines and the label "piecewise linear interpolation." The lower part shows the processed unified time axis running samples, where all variables are aligned to the same sampling time, forming a regular sampling period sequence. In the actual operation of the SCR system, the sampling period of the distributed control system is typically on the order of seconds, the response period of the flue gas online monitoring system may reach the order of minutes, and the feedback signals of the actuators have their own independent refresh frequencies. If this sampling asynchronicity is not addressed, it will lead to time misalignment in subsequent rate of change calculations and state estimations. This method employs piecewise linear interpolation with monotonicity constraints, which, while ensuring data continuity, suppresses non-physical oscillations that may be introduced during the interpolation process, enabling precise alignment of multi-source data on a unified time axis.

[0094] S123: Based on the unified time axis running sample matrix, perform missing point filling processing to obtain a complete running sample matrix; the missing point filling processing adopts a combination strategy of forward hold and backward hold with boundary constraints. The boundary constraints are obtained by statistically analyzing the upper and lower limits of physical quantities under the same operating condition type in the multi-source original running data set. The operating condition type is jointly determined by the unit load range and the flue gas flow range to ensure that the filling value does not exceed the operating range. The final output complete running sample matrix meets the requirements of data continuity and standardization for subsequent outlier removal, operating condition segmentation and other processing.

[0095] S13: Perform outlier removal based on the complete running sample matrix to obtain a cleaned running sample matrix;

[0096] Furthermore, the outlier removal process includes the following steps:

[0097] S131: Calculate the robust residual sequence of variables based on the complete running sample matrix to obtain the robust residual matrix of variables; the robust residual sequence of variables is obtained by subtracting the original value from the mean value filtering result in the sliding window. The length of the sliding window is determined by the sampling period and the nominal transmission time of flue gas from the SCR inlet to the outlet, so as to ensure that the robust residual sequence of variables can cover the error pattern dominated by transmission delay.

[0098] S132: Calculate the anomaly detection threshold set based on the robust residual matrix of the variables to obtain the anomaly detection threshold table; the anomaly detection threshold set is calculated separately for each variable, and the calculation method of the anomaly detection threshold is as follows: sort the absolute values ​​of the robust residual sequence of the variables and take the upper quantile as the anomaly detection threshold. The upper quantile is determined by back-calculation of the proportion of steady-state operation segments within the regulatory assessment period, so that the anomaly detection threshold is adaptively updated with the measurement noise level.

[0099] S133: Based on the anomaly discrimination threshold table, perform anomaly point clearing and secondary filling processing on the complete running sample matrix to obtain the cleaned running sample matrix; the anomaly point clearing and secondary filling processing is to first set the sample points exceeding the anomaly discrimination threshold as missing, and then call the missing point filling processing of S122 to generate replacement values, so as to ensure that the cleaned running sample matrix maintains continuity for subsequent identification calculations.

[0100] S14: Perform segmented processing based on the cleaning operation sample matrix to obtain the working condition feature sample set;

[0101] Furthermore, the segmented processing of the operating conditions includes the following steps:

[0102] S141: Calculate the temperature change rate matrix and flow rate change rate sequence based on the cleaning operation sample matrix to obtain the change rate feature matrix; the temperature change rate matrix is ​​the difference result of the inlet flue gas temperature and outlet flue gas temperature of each catalyst layer on a unified time axis, and the flow rate change rate sequence is the difference result of the inlet flue gas flow rate of the SCR reactor. The difference adopts the central difference and adds a smoothing constraint to suppress the amplification of measurement noise.

[0103] S142: Calculate the steady-state discrimination threshold and the rapid temperature change discrimination threshold based on the rate of change feature matrix to obtain segmented threshold parameters. The steady-state discrimination threshold is calculated from the lower quantile of the absolute value distribution in the rate of change feature matrix, and the rapid temperature change discrimination threshold is calculated from the upper quantile of the absolute value distribution in the rate of change feature matrix. The lower and upper quantiles are obtained statistically from the frequency of changes in the unit's automatic power generation control peak-shaving command, so that the segmented threshold parameters match the peak-shaving intensity. (See also...) Figure 3This is a schematic diagram of the segmented processing of operating conditions and the discrimination of steady-state rapid temperature change provided in the embodiments of this application. As shown in the figure, the horizontal axis is time, and the vertical axis is the rate of temperature change. The blue curve in the figure represents the evolution trajectory of the rate of temperature change over time. Two sets of horizontal threshold lines are set in the figure: the red dashed line represents the rapid temperature change discrimination threshold, and the green dashed line represents the steady-state discrimination threshold. When the rate of temperature change falls within the green threshold range, the corresponding time interval is marked as a steady-state operating condition segment and filled with a green background; when the rate of temperature change exceeds the red threshold, the corresponding interval is marked as a rapid temperature change operating condition segment and filled with a red background. The figure marks two typical operating conditions: "rapid heating" and "rapid cooling". In the scenario where thermal power units participate in automatic power generation control and peak shaving, the rapid rise and fall of load will inevitably lead to drastic changes in the flue gas temperature in the furnace. This temperature change process will cause a significant change in the ammonia adsorption and desorption behavior on the catalyst surface. By explicitly segmenting the operating data into two types of conditions—steady-state and rapid temperature change—subsequent parameter identification can use steady-state data to estimate the equilibrium ammonia coverage rate parameter and ammonia saturation storage capacity parameter, and use rapid temperature change data to estimate the storage capacity temperature slope parameter and dynamic response time parameter, thereby avoiding misjudging transient adsorption and desorption behavior caused by temperature gradient as conventional measurement noise.

[0104] S143: Based on the segmented threshold parameter, the cleaning operation sample matrix is ​​divided into segments to obtain a set of steady-state operating condition segments and a set of rapid temperature change operating condition segments. The set of steady-state operating condition segments is a set of time intervals in which both the temperature change rate and the flow rate change rate do not exceed the steady-state discrimination threshold. The set of rapid temperature change operating condition segments is a set of time intervals in which the temperature change rate exceeds the rapid temperature change discrimination threshold and the duration exceeds the minimum segment duration. The minimum segment duration is determined by the sampling period and the minimum discernible dynamic duration of the ammonia injection actuator.

[0105] S144: Based on the set of steady-state operating conditions and the set of rapid temperature-changing operating conditions, perform sample labeling processing on the cleaning operation sample matrix to obtain the operating condition feature sample set; the operating condition feature sample set is a set of labeled samples, and each sample includes at least the operating condition label, unit load, flue gas flow rate, inlet nitrogen oxide concentration, outlet nitrogen oxide concentration, total ammonia injection flow rate, ammonia slip concentration, inlet flue gas temperature of each catalyst layer, and outlet flue gas temperature of each catalyst layer.

[0106] S15: Based on the working condition feature sample set and the steady-state working condition segment set, perform steady-state ammonia storage parameter identification processing to obtain the steady-state ammonia storage parameter table;

[0107] Furthermore, the steady-state ammonia storage parameter identification process includes the following steps:

[0108] S151: Based on the set of steady-state operating condition segments, extract a subset of steady-state equilibrium samples from the set of operating condition feature samples to obtain a steady-state equilibrium sample set; the steady-state equilibrium sample set is a set of samples sampled within the same steady-state operating condition segment and satisfying that the fluctuations of inlet nitrogen oxide concentration and flue gas flow rate do not exceed the steady-state discrimination threshold, so as to ensure that the steady-state equilibrium sample set can be used to approximately represent the equilibrium state of ammonia coverage on the catalyst surface.

[0109] S152: Calculate the ammonia equivalent demand sequence based on the steady-state equilibrium sample set to obtain the steady-state ammonia equivalent demand matrix; the ammonia equivalent demand sequence is obtained by converting the difference between the inlet nitrogen oxide concentration and the outlet nitrogen oxide concentration. The conversion adopts the stoichiometric equivalent relationship between nitrogen oxides and ammonia and combines the flue gas flow rate to convert the concentration difference into equivalent flow rate, so as to form a comparison benchmark of the same dimension as the total ammonia injection flow rate.

[0110] S153: Calculate the steady-state ammonia balance residual sequence based on the steady-state ammonia equivalent demand matrix and the steady-state balance sample set to obtain the steady-state ammonia balance residual set; the steady-state ammonia balance residual sequence is obtained by subtracting the ammonia equivalent demand from the total ammonia injection flow rate and deducting the escaped ammonia equivalent flow rate converted from the ammonia escape concentration; the steady-state ammonia balance residual set is used to characterize the difference in ammonia quantity caused by catalyst storage and local mixing inhomogeneity under steady-state conditions.

[0111] S154: Based on the steady-state ammonia balance residual set and the inlet flue gas temperature of each catalyst layer, a bin-based statistical processing is performed to obtain the balance ammonia coverage parameter and the ammonia saturated storage capacity parameter. The bin-based statistical processing constructs a two-dimensional bin index according to the flue gas space velocity range and temperature range. The flue gas space velocity is calculated from the inlet flue gas flow rate of the SCR reactor and the geometric volume of the catalyst. The balance ammonia coverage parameter is defined as a combined index of the sign consistency and amplitude center statistics of the steady-state ammonia balance residual within the corresponding bin, mapped to a dimensionless value within the range of 0 to 1. The ammonia saturated storage capacity parameter is defined as the statistical measure of the critical ammonia equivalent increment at which the ammonia escape concentration begins to rise continuously when the total ammonia injection flow rate is gradually increased within the corresponding bin. The critical ammonia equivalent increment is extracted from the sample sequence in the steady-state balance sample set that satisfies the condition that the inlet nitrogen oxide concentration remains essentially unchanged. In a preferred embodiment, the balance ammonia coverage parameter... The specific calculation method is as follows: Let the steady-state ammonia balance residual sequence in the corresponding sub-tank be... Define the symbol consistency index With amplitude center statistics They are respectively:

[0112]

[0113]

[0114] in For symbolic functions, For median operations; , A value close to positive indicates that the ammonia injection rate is systematically high under steady-state conditions. A value close to negative indicates a systematic underestimation; let the ammonia saturation storage capacity parameter in this compartment be... The equilibrium ammonia coverage parameter is:

[0115]

[0116] in This is a truncation function that restricts the result to an interval. within; when For positive and When the value is positive, it indicates that the stored ammonia is close to saturation under steady state. Approaching 1; when negative and A negative value indicates that the stored ammonia is far below saturation under steady-state conditions. Approaching 0.

[0117] In a preferred embodiment, the ammonia saturation storage capacity parameter The extraction method is as follows: From the steady-state equilibrium sample set, sample sequences that satisfy the condition that the absolute value of the inlet nitrogen oxide concentration change rate does not exceed the steady-state discrimination threshold are selected, sorted in ascending order by the total ammonia injection flow rate, and the difference sequence of ammonia escape concentration between adjacent samples is calculated. When three or more consecutive difference values ​​are positive and their mean exceeds 1.5 times the standard deviation of the difference sequence, the difference between the total ammonia injection flow rate and the ammonia equivalent demand corresponding to the sample point that first meets the condition is taken as the critical ammonia equivalent increment for that extraction. The median of all extraction results that meet the condition within the same compartment is taken as the ammonia saturation storage capacity parameter of that compartment. .

[0118] S16: Based on the working condition feature sample set and the rapid temperature change working condition segment set, perform dynamic parameter identification processing for variable temperature ammonia storage to obtain the dynamic parameter table for variable temperature ammonia storage.

[0119] Furthermore, the dynamic parameter identification process for variable-temperature ammonia storage includes the following steps:

[0120] S161: Based on the rapid temperature change working condition segment set, extract the temperature change response sample subset from the working condition feature sample set to obtain the temperature change response sample set; the temperature change response sample set is a sample set that satisfies that the total ammonia injection flow rate remains continuous at the beginning of the segment and that the ammonia injection valve opening is not saturated, so as to reduce the interference of actuator saturation on dynamic identification.

[0121] S162: Calculate the temperature gradient driven residual sequence based on the variable temperature response sample set to obtain the temperature gradient driven residual set; the temperature gradient driven residual sequence is obtained by subtracting the influence of the equilibrium ammonia coverage parameter corresponding to the same air velocity and temperature in the steady-state operating condition segment set from the steady-state ammonia equilibrium residual sequence, so as to separate the storage offset caused by temperature change from the steady-state bias.

[0122] S163: Based on the temperature gradient-driven residual set and temperature change rate matrix, recursive least squares identification processing is performed to obtain the storage capacity temperature slope parameter and dynamic response time parameter. The recursive least squares identification processing adopts a recursive update method with a forgetting factor. The forgetting factor is determined by the assumption of slow change in catalyst deactivation rate and is consistent with the update cycle. The storage capacity temperature slope parameter is defined as the equivalent slope of the change in ammonia saturated storage capacity parameter caused by a unit temperature change. The dynamic response time parameter is defined as the characteristic time scale of the convergence of the ammonia coverage state quantity from the old equilibrium to the new equilibrium after a step change in temperature. The dynamic response time parameter is obtained by minimizing the weighted loss of the ammonia escape concentration prediction error and the outlet nitrogen oxide concentration prediction error.

[0123] S17: Construct a catalyst temperature storage index tree based on the steady-state ammonia storage parameter table and the variable-temperature ammonia storage dynamic parameter table to obtain the catalyst temperature storage index tree;

[0124] Furthermore, the construction of the catalyst temperature storage index tree includes the following steps:

[0125] S171: Based on the structural parameters of the SCR reactor and the catalyst layer arrangement parameters, a catalyst spatial unit set is divided to obtain a catalyst spatial unit index table; the catalyst spatial unit set is jointly encoded by the catalyst layer number and the segment number along the flue gas flow direction, and the segment number along the flue gas flow direction is jointly determined by the effective length of the reactor and the flue gas transmission discreteness. The catalyst spatial unit index table is used as a node positioning key in the subsequent ammonia storage dynamic response diagram.

[0126] S172: Based on the catalyst spatial unit index table, perform spatial binding processing on the steady-state ammonia storage parameter table and the variable-temperature ammonia storage dynamic parameter table to obtain a spatially bound ammonia storage parameter set; the spatial binding processing maps each catalyst spatial unit set to the corresponding temperature range and flue gas space velocity range, and generates a parameter record for each mapping result. The parameter record includes at least the ammonia saturated storage capacity parameter, the equilibrium ammonia coverage parameter, the storage capacity temperature slope parameter, and the dynamic response time parameter.

[0127] S173: A confidence parameter and deactivation coefficient are calculated based on a spatially bound ammonia storage parameter set to obtain a confidence-deactivation parameter set. The confidence parameter is calculated by combining the number of valid samples corresponding to the parameter record with the robustness scale of the fitting residual. The number of valid samples is obtained by counting samples that meet the segmented threshold parameter constraint and are not removed by outliers. The deactivation coefficient is calculated by the degradation trend of the outlet nitrogen oxide concentration under the same total ammonia injection flow rate and the same inlet nitrogen oxide concentration conditions within a long-term window. The length of the long-term window is determined by the time scale of catalyst performance drift to ensure that the deactivation coefficient can reflect the impact of catalyst activity decay on ammonia storage behavior. In a preferred embodiment, the confidence parameter... The calculation method is as follows: Let the number of valid samples corresponding to the parameter record be... The robustness metric for the fitted residuals is (Obtained by multiplying the median absolute deviation of the fitted residual sequence by 1.4826), the preset minimum effective sample size is... The preset benchmark residual scale is ,but:

[0128]

[0129] in ; Take twice the median of the number of valid samples in each bin. Take the median of the robustness scale of all binned fit residuals; when the number of effective samples is larger and the fit residuals are smaller... The closer the value is to 1, the higher the reliability of the parameter record.

[0130] In a preferred embodiment, the deactivation coefficient The calculation method is as follows: Within a long-term window, steady-state samples are selected that meet the criteria of total ammonia injection flow deviation not exceeding the steady-state discrimination threshold and inlet nitrogen oxide concentration deviation not exceeding the steady-state discrimination threshold. A univariate linear regression is performed with cumulative operating time as the independent variable and outlet nitrogen oxide concentration as the dependent variable to obtain the regression slope. (Unit: concentration / time); Assume the baseline outlet nitrogen oxide concentration at the initial catalyst installation is... The cumulative running time from the initial installation to the center time of the long-term window is ,but:

[0131]

[0132] in ; This indicates that the catalyst has not deactivated. A smaller value indicates a more severe degradation of catalyst activity; the deactivation coefficient is used in a multiplicative reduction manner when applied to the ammonia saturated storage capacity parameter, i.e., the equivalent ammonia saturated storage capacity is... .

[0133] S174: Based on the spatially bound ammonia storage parameter set and the reliability deactivation parameter set, a tree node record is constructed to obtain the catalyst temperature storage index tree node set. The catalyst temperature storage index tree is a hierarchical tree structure. The root node corresponds to the entire SCR reactor, the first-level nodes correspond to the catalyst layer number, the second-level nodes correspond to the temperature range, the third-level nodes correspond to the flue gas space velocity range, and the leaf nodes store tree node records. Each tree node record includes at least the ammonia saturation storage capacity parameter, equilibrium ammonia coverage parameter, storage capacity temperature slope parameter, dynamic response time parameter, reliability parameter, and deactivation coefficient. The code in the catalyst spatial unit index table is used as the index key to support rapid retrieval by temperature and space velocity during operation. See also... Figure 4 This is a schematic diagram of the catalyst temperature storage index tree provided in this application embodiment. As shown in the figure, the tree structure adopts a top-down hierarchical organization: the root node represents the entire SCR reactor; the first-level index is divided according to the catalyst layer number, with "catalyst layer N" shown as an example to illustrate the selected path; the second-level index is divided according to temperature range, mapping different temperature ranges to corresponding branches; the third-level index is divided according to flue gas space velocity range, further refining the search conditions. The bottom-level leaf nodes store complete tree node records, including six key parameters: ammonia saturation storage capacity parameter, equilibrium ammonia coverage parameter, storage capacity temperature slope parameter, dynamic response time parameter, confidence parameter, and deactivation coefficient. The hierarchical label on the left side of the figure clearly defines the meaning of each level of the index. During the control cycle, when it is necessary to infer the ammonia storage behavior of a certain catalyst spatial unit under specific temperature and space velocity conditions, the system can quickly locate the corresponding leaf node through the catalyst layer number, current temperature value, and current space velocity value, and directly read the pre-identified parameter records. This hierarchical index structure reduces retrieval complexity from linear to logarithmic, ensuring response speed in real-time control scenarios. At the same time, by incorporating reliability parameters and deactivation coefficients as components of parameter records, the control system can simultaneously acquire information on parameter reliability and catalyst activity decay.

[0134] S175: Generate a rolling update sample queue based on the operating condition feature sample set to obtain the index tree update input queue; the index tree update input queue is a sample queue enqueued by time, and the index tree update input queue is used to trigger the incremental update of the catalyst temperature storage index tree in the online adaptive write-back process of S4, so as to ensure that the catalyst temperature storage index tree remains consistent with the catalyst deactivation and load structure changes.

[0135] Specifically, S1 explicitly separates steady-state and rapid temperature-changing operating conditions using a set of operating condition characteristic samples. This allows subsequent parameter identification to estimate the equilibrium ammonia coverage parameter and ammonia saturation storage capacity parameter using steady-state data, and to estimate the storage capacity temperature slope parameter and dynamic response time parameter using rapid temperature-changing data. This avoids mistaking transient adsorption and desorption behavior caused by temperature gradients as routine measurement noise or transmission delay errors. The catalyst temperature storage index tree structurally encodes the catalyst layer, temperature range, and flue gas space velocity range in a unified manner, and incorporates reliability parameters and deactivation coefficients as components of the parameter records. This ensures that searches within the control cycle not only yield ammonia storage capacity but also the sensitivity of ammonia storage capacity to temperature changes and the reliability of parameters. This provides S2 with a directly callable calculable object for inferring adsorption and desorption fluxes during rapid temperature changes. Without the catalyst temperature storage index tree, subsequent steps cannot convert the temperature change rate into a quantifiable change in stored ammonia. Ammonia injection feedforward control can only remain at the stoichiometric level, unable to explain the systematic deviations of sudden increases in ammonia escape during temperature rises and short-term exceedances of outlet nitrogen oxides during temperature drops.

[0136] S2: Based on the catalyst temperature storage index tree and real-time temperature change rate, the ammonia adsorption and desorption flux distribution is inferred to obtain the ammonia storage dynamic response map;

[0137] Further, S2 includes the following steps:

[0138] S21: Based on the unified time axis running sample matrix, read the running sample vector at the current moment from the real-time acquisition window to obtain the real-time running sample vector; the real-time running sample vector includes at least the current unit load, the current SCR reactor inlet flue gas flow rate, the current inlet nitrogen oxide concentration, the current outlet nitrogen oxide concentration, the current total ammonia injection flow rate, the current ammonia slip concentration, the inlet flue gas temperature of each catalyst layer, and the outlet flue gas temperature of each catalyst layer.

[0139] S22: Based on the real-time running sample vector and the historical samples corresponding to the preset number of review steps in the unified time axis running sample matrix, differential smoothing is performed to obtain the real-time temperature change rate vector and the real-time flow change rate value; the preset number of review steps is determined by the sampling period and the noise main frequency bandwidth of the temperature measurement point. The differential smoothing process adopts a constrained sliding least squares linear fitting. The constraint is that the fitting slope must not exceed the upper quantile of the corresponding change rate distribution in the clean running sample matrix, so as to avoid distortion of the temperature change rate due to single-point jump.

[0140] S23: Short-time temperature prediction processing is performed based on the real-time temperature change rate vector and the real-time operating sample vector to obtain a short-time temperature prediction sequence. The short-time temperature prediction processing adopts a recursive extrapolation method with physical boundaries. The physical boundaries are jointly determined by the catalyst's allowable operating temperature range and the boiler load ramp-up rate range. The short-time temperature prediction sequence is a discrete-time sequence and covers a preset number of sampling periods for prediction steps. In a preferred embodiment, the short-time temperature prediction processing adopts a first-order linear recursive extrapolation method with physical boundaries. Specifically, starting with the temperature value of each catalyst layer at the current moment, and using the corresponding component in the real-time temperature change rate vector as the slope, the temperature prediction values ​​for each future prediction moment are generated by progressively recursively generating the predicted temperature values ​​according to the sampling period. In each recursive step, if the predicted value exceeds the upper or lower limit of the catalyst's allowable operating temperature range, the predicted value is truncated to the corresponding boundary value, and the slope of the subsequent recursive steps is set to zero. If the absolute value of the corresponding component in the real-time temperature change rate vector, after being converted into a unit-time load change rate, exceeds the upper limit of the boiler load ramp-up rate range, the temperature change rate is truncated to the equivalent temperature change rate corresponding to the upper limit of the load ramp-up rate.

[0141] S24: Calculate the flue gas space velocity prediction sequence based on the short-time temperature prediction sequence and the real-time running sample vector to obtain the space velocity prediction sequence; the flue gas space velocity prediction sequence is obtained by converting the flue gas flow rate at the SCR reactor inlet with the catalyst geometric volume, and combined with the real-time flow rate change value for short-time extrapolation to cover the preset prediction steps.

[0142] S25: Based on the short-time temperature prediction sequence, space velocity prediction sequence, and catalyst spatial unit index table, perform index retrieval processing to obtain a set of local ammonia storage parameter vectors; the index retrieval processing is to locate the leaf node in the catalyst temperature storage index tree for each catalyst spatial unit index key in the catalyst spatial unit index table using the temperature value corresponding to the spatial unit in the short-time temperature prediction sequence and the space velocity value in the space velocity prediction sequence, and read the ammonia saturation storage capacity parameter, equilibrium ammonia coverage parameter, storage capacity temperature slope parameter, dynamic response time parameter, confidence parameter, and deactivation coefficient stored in the leaf node, thereby forming a set of local ammonia storage parameter vectors arranged by spatial unit.

[0143] S26: Based on the local ammonia storage parameter vector set and the short-term temperature prediction sequence, perform equilibrium storage capacity update processing to obtain the target equilibrium ammonia storage capacity sequence; wherein, the target equilibrium ammonia storage capacity sequence is obtained by mapping the ammonia saturated storage capacity parameter and the storage capacity temperature slope parameter to the short-term temperature prediction sequence, and the ammonia saturated storage capacity parameter is reduced and corrected according to the deactivation coefficient so that the target equilibrium ammonia storage capacity sequence reflects the characteristic of the upper limit of ammonia that can be stored after catalyst aging. The ammonia saturated storage capacity parameter Ωsat is multiplicatively reduced and corrected according to the deactivation coefficient δ, that is, the reduced equivalent ammonia saturated storage capacity is δ⋅Ωsat.

[0144] S27: Based on the target equilibrium ammonia storage sequence and the equilibrium ammonia coverage parameter, perform initial ammonia coverage state quantity estimation processing to obtain an initial ammonia coverage state quantity set; wherein, the initial ammonia coverage state quantity is defined as the ratio of the actual stored ammonia in the catalyst space unit at the current moment to the target equilibrium ammonia storage at the current temperature. The initial ammonia coverage state quantity set is obtained by inputting the equilibrium ammonia coverage parameter and the current ammonia escape concentration deviation into a state constraint estimation algorithm. The state constraint estimation algorithm adopts a recursive correction method with upper and lower bounds, the upper and lower bounds being 0 to 1. In a preferred embodiment, the state constraint estimation algorithm specifically adopts a recursive Bayesian correction method with truncation constraints, that is: using the equilibrium ammonia coverage parameter As a priori estimate of the ammonia coverage state quantity, the measured value of the current ammonia escape concentration is compared with the value based on... The difference in predicted ammonia escape concentrations is used as the observation residual. The prior estimate is then additively corrected using a proportionally corrected gain. If the corrected estimate exceeds the interval... The value is then truncated to the corresponding boundary value to obtain the initial ammonia coverage state quantity; the proportional correction gain is determined by the confidence parameter of the current bin. Inverse proportion determination, that is The higher the value, the smaller the correction gain, indicating greater confidence in the prior parameters.

[0145] S28: Based on the initial ammonia coverage state set, the target equilibrium ammonia storage sequence, and the dynamic response time parameter, a dynamic convergence calculation is performed to obtain the ammonia coverage state prediction sequence. The dynamic convergence calculation uses the dynamic response time parameter as a convergence speed constraint, and performs discrete iterative updates on each catalyst spatial unit within a preset number of prediction steps, causing the ammonia coverage state prediction sequence to converge from the initial ammonia coverage state to a new equilibrium state induced by the target equilibrium ammonia storage sequence. In a preferred embodiment, the discrete iterative update adopts a forward Euler discretization method, specifically: Let the first... The predicted ammonia coverage state value at each predicted time is The corresponding target balanced ammonia coverage rate is (Obtained by dividing the target balanced ammonia storage capacity by the reduced equivalent ammonia saturation storage capacity), the dynamic response time parameter is: The sampling period is ,but:

[0146]

[0147] in This represents the initial ammonia coverage state quantity. , The preset prediction step number; if during the iteration process Exceeding the range Then it is truncated to the corresponding boundary value.

[0148] S29: Based on the ammonia coverage state quantity prediction sequence and the target equilibrium ammonia storage quantity sequence, perform storage ammonia change calculation processing to obtain a set of local storage ammonia change prediction sequences; the local storage ammonia change prediction sequence is the difference result of the storage ammonia quantity sequence obtained by multiplying the target equilibrium ammonia storage quantity sequence and the ammonia coverage state quantity prediction sequence at adjacent time points, the sign of the local storage ammonia change prediction sequence is positive to indicate the desorption trend, and the sign of the local storage ammonia change prediction sequence is negative to indicate the adsorption trend.

[0149] S210: Based on the local storage ammonia change prediction sequence set and sampling period, perform flux conversion processing to obtain the ammonia adsorption and desorption flux distribution; the flux conversion processing is to divide the local storage ammonia change prediction sequence by the sampling period and combine it with the equivalent volume weight of the catalyst spatial unit to obtain the storage ammonia release flux or storage ammonia absorption flux per unit time. The ammonia adsorption and desorption flux distribution is stored with the catalyst spatial unit index key and the prediction time index as a two-dimensional index.

[0150] S211: Based on the ammonia adsorption and desorption flux distribution and the catalyst spatial unit index table, a transport delay mapping process is performed to obtain a unit transport delay table; the unit transport delay mapping process sorts the catalyst spatial units into segments according to the flue gas flow direction, and calculates the discrete delay steps of the flue gas passing through the segment based on the equivalent length of each segment and the space velocity prediction sequence. The unit transport delay table is used to determine the time offset of the edge when constructing the ammonia storage dynamic response diagram.

[0151] S212: A dynamic response diagram for ammonia storage is constructed based on the flux distribution of ammonia adsorption and desorption, the predicted sequence of ammonia coverage state variables, and the unit transport delay table. The dynamic response diagram is a directed graph data structure. The nodes of the dynamic response diagram are jointly encoded by the catalyst spatial unit index key and the predicted time index. Node attributes include at least the predicted temperature value, predicted space velocity value, predicted ammonia coverage state variable value, and ammonia adsorption and desorption flux values. The directed edges of the dynamic response diagram connect upstream and downstream nodes along the flue gas flow direction, and the flux influence of the upstream nodes is mapped to the corresponding predicted time index of the downstream nodes according to the unit transport delay table. (See also...) Figure 5This is a schematic diagram of the dynamic response of ammonia storage provided in an embodiment of this application. As shown in the figure, the diagram adopts a two-dimensional directed graph structure. The horizontal axis is the prediction time axis (extending from the current time T to the future T+n), and the vertical axis is the flue gas flow direction (extending from the upstream catalyst layer to the downstream catalyst layer). The rectangular nodes in the figure represent the state units of a specific catalyst layer at a specific prediction time. The nodes are marked with the flux generation information driven by the rate of temperature change and the iterative update process of the ammonia coverage state. The blue solid arrows indicate the convergence evolution process of the ammonia coverage state along the time axis within the same catalyst layer, and the red dashed arrows indicate the transmission mapping relationship of flux along the flue gas flow direction from the upstream node to the downstream node. The "unit transmission delay" marked next to the arrows indicates the number of discrete time steps required for the flue gas to cross adjacent catalyst layers. The legend box further clarifies the physical meaning of the two types of arrows. During the rapid temperature change phase, the desorption flux or adsorption flux generated by the upstream catalyst layer does not immediately affect the outlet measuring point, but needs to undergo the flue gas transmission delay before reaching the downstream and finally being reflected in the outlet nitrogen oxide concentration and ammonia escape concentration. The dynamic response diagram of ammonia storage explicitly encodes the relationship between spatially distributed flux and time delay as directed edges, enabling the control system to distinguish the differences in storage behavior caused by different temperature change trends under the same temperature value, and providing a quantifiable predictive basis for ammonia injection feedforward compensation.

[0152] S213: Perform rolling pruning on the ammonia storage dynamic response graph based on the real-time running sample vector to obtain a rolling ammonia storage dynamic response graph; the rolling pruning process is to remove the nodes and edges corresponding to the prediction time index that are earlier than the current time in the ammonia storage dynamic response graph in each control cycle, and supplement the nodes and edges corresponding to the new prediction time index at the end according to the short-time temperature prediction sequence, thereby forming a rolling ammonia storage dynamic response graph aligned with the control cycle.

[0153] Specifically, S2 transforms the static parameters searchable in the catalyst temperature storage index tree into ammonia adsorption and desorption flux distribution driven by the rate of temperature change. It then uses a unit transmission delay table to explicitly encode the time impact of local flux on downstream outlet measuring points into the rolling ammonia storage dynamic response diagram. This allows the control system to distinguish storage behavior differences caused by different temperature change trends at the same temperature value. During the rapid temperature rise phase, the predicted sequence of local stored ammonia changes shows a desorption trend. If the delay relationship of desorption flux propagation along the flow direction is not expressed in the rolling ammonia storage dynamic response diagram, the ammonia injection command can only be corrected by the lag feedback of outlet nitrogen oxide concentration and ammonia escape concentration, easily leading to a sudden increase in ammonia escape before the feedback occurs. During the rapid temperature fall phase, the predicted sequence of local stored ammonia changes shows an adsorption trend. If there is a lack of forward estimation of adsorption flux, the ammonia injection feedforward will still supply ammonia according to stoichiometry, resulting in effective reactive ammonia being occupied by storage, thus causing short-term exceedances of outlet nitrogen oxide concentration that are difficult to suppress in time. By using the rolling dynamic response diagram of ammonia storage, S3 can directly perform equivalent calculations on the release and absorption of stored ammonia in future control cycles, thus realizing a feedforward compensation entry for temperature gradient.

[0154] S3: Calculate the equivalent stored ammonia flow rate based on the ammonia storage dynamic response diagram and superimpose it with the theoretical ammonia injection rate to generate a distribution-type ammonia injection feedforward compensation sequence;

[0155] Further, S3 includes the following steps:

[0156] S31: Based on the inlet nitrogen oxide concentration, outlet nitrogen oxide concentration target value in the real-time running sample vector and the current SCR reactor inlet flue gas flow rate, perform chemometric conversion processing to obtain the theoretical ammonia injection rate; the outlet nitrogen oxide concentration target value is generated by the emission control strategy module according to emission limits, load range and ammonia slip constraints; the chemometric conversion processing converts the difference between the inlet nitrogen oxide concentration and the outlet nitrogen oxide concentration target value into the required ammonia equivalent and converts it into the ammonia injection total flow rate setting benchmark; the theoretical ammonia injection rate is a continuous value and has engineering units.

[0157] S32: Based on the outlet nitrogen oxide concentration and ammonia escape concentration in the real-time running sample vector, feedback control calculation is performed to obtain the feedback-corrected ammonia injection amount. The feedback control calculation adopts a dual feedback structure. The first feedback generates an ammonia increase correction amount based on the deviation between the outlet nitrogen oxide concentration and the target value of the outlet nitrogen oxide concentration. The second feedback generates an ammonia decrease correction amount based on the deviation between the ammonia escape concentration and the ammonia escape constraint value. Anti-saturation and anti-integral accumulation constraints are applied to the two correction amounts to ensure that the feedback-corrected ammonia injection amount changes continuously within the adjustable range of the actuator. In a preferred embodiment, the feedback control calculation adopts a dual feedback structure, specifically: the first feedback uses a proportional-integral controller with integral anti-saturation, taking the deviation between the outlet nitrogen oxide concentration and the target value of the outlet nitrogen oxide concentration as input, and outputs an ammonia increase correction amount; the second feedback uses a pure proportional controller, taking the deviation between the ammonia escape concentration and the ammonia escape constraint value as input, and outputs an ammonia decrease correction amount when the ammonia escape concentration exceeds the ammonia escape constraint value, otherwise the output is zero. The anti-saturation constraint is specifically implemented using a conditional integration method. That is, when the output of the first feedback controller reaches the upper or lower limit of the output amplitude, the accumulation and updating of the integral term is paused until the output leaves the saturation range, at which point integral accumulation resumes. Specifically, the anti-integral accumulation constraint is as follows: when both the ammonia increase correction and the ammonia decrease correction are non-zero, the integral term of the first feedback controller decays in the opposite direction by an equal amount based on the absolute value of the ammonia decrease correction, to avoid the continuous accumulation of the integral term due to contradictory directions of the two corrections.

[0158] S33: Based on the rolling ammonia storage dynamic response graph, traverse the directed edges and accumulate flux contributions to obtain the equivalent storage ammonia flow sequence; the method of traversing the directed edges and accumulating flux contributions is as follows: traverse the directed edges pointing to the outlet direction in the rolling ammonia storage dynamic response graph from near to far according to the predicted time index, map the ammonia adsorption and desorption flux values ​​to the outlet equivalent time index according to the unit transmission delay table and accumulate them to obtain the outlet equivalent flux sequence, and then convert the outlet equivalent flux sequence into the equivalent storage ammonia flow sequence according to the engineering unit of the total ammonia injection flow; the components greater than 0 in the equivalent storage ammonia flow sequence indicate that there is equivalent ammonia supply from catalyst storage in the future period, and the components less than 0 indicate that there is equivalent ammonia deficiency caused by catalyst adsorption in the future period.

[0159] S34: Perform feedforward compensation synthesis processing based on the equivalent stored ammonia flow rate sequence and the theoretical ammonia injection rate to obtain the ammonia injection feedforward compensation sequence; the feedforward compensation synthesis processing is to perform algebraic superposition of the theoretical ammonia injection rate and the equivalent stored ammonia flow rate sequence at each prediction time index of the preset prediction steps and apply safety boundary constraints. The safety boundary is determined by the maximum ammonia supply capacity and the minimum stable ammonia injection capacity of the ammonia injection system, thereby generating an ammonia injection feedforward compensation sequence containing the ammonia injection set values ​​for multiple future sampling periods.

[0160] S35: Calculate the feedforward compensation weight coefficient and the feedback correction weight coefficient based on the real-time temperature change rate vector and the unit load change rate to obtain a set of weight coefficients; the feedforward compensation weight coefficient is obtained by normalizing the mean absolute value of the real-time temperature change rate vector and the absolute value of the unit load change rate, and the feedback correction weight coefficient is obtained by subtracting the feedforward compensation weight coefficient from 1. The upper and lower limits of normalization are determined by the quantiles of the corresponding change rate distribution in the cleaning operation sample matrix to avoid the weight coefficient set being overly sensitive to single peak changes.

[0161] S36: Based on the ammonia injection feedforward compensation sequence, the feedback correction ammonia injection amount, and the set of weight coefficients, perform composite setpoint fusion processing to obtain a composite ammonia injection setpoint sequence; the composite setpoint fusion processing is to superimpose the ammonia injection feedforward compensation amount weighted by the feedforward compensation weight coefficient at the first element of the ammonia injection feedforward compensation sequence corresponding to the current control cycle, and superimpose the feedback correction ammonia injection amount weighted by the feedback correction weight coefficient to form the current ammonia injection setpoint, while retaining the ammonia injection feedforward compensation sequence elements corresponding to the other prediction time indices as the look-ahead setpoints at the tail of the queue, so that S4 can perform queue-style distribution.

[0162] S37: Based on the composite ammonia injection setpoint sequence and the layered ammonia injection flow constraint table, an allocation calculation process is performed to obtain an allocation-type ammonia injection feedforward compensation sequence. The layered ammonia injection flow constraint table is jointly determined by the ammonia injection grid arrangement, the flow characteristics of valves in each layer, and the uniformity requirements of the ammonia-nitrogen molar ratio. The allocation calculation process adopts a constrained proportional allocation and superimposes a correction term based on the temperature deviation of each layer. The temperature deviation is calculated from the inlet flue gas temperature of each catalyst layer and the average temperature of the entire layer, so that the allocation-type ammonia injection feedforward compensation sequence can take into account the differences in interlayer reaction activity.

[0163] Specifically, S3 converts the changes in storage flux in the dynamic response diagram of rolling ammonia storage into an equivalent storage ammonia flow sequence. This allows the desorption ammonia supply caused by rapid temperature rise to be identified and used to reduce the ammonia injection setpoint before the outlet ammonia escape concentration increases, and the adsorption occupancy caused by rapid temperature drop to be identified and used to increase the ammonia injection setpoint before the outlet nitrogen oxide concentration exceeds the limit. Through adaptive fusion of the feedforward compensation weight coefficient and the feedback correction weight coefficient, the composite ammonia injection setpoint sequence increases the dependence on model prediction when the temperature change rate and load change rate are large, thereby offsetting the systematic deviation caused by storage in advance; when the change rate is small and the reliability of the measurement point feedback is high, it increases the dependence on the feedback correction ammonia injection amount, thereby suppressing long-term mismatch and the accumulation of modeling errors. The distributed ammonia injection feedforward compensation sequence further applies the total compensation to the stratification constraints and inter-layer temperature differences, so that the compensation is not only correct in terms of total amount, but also avoids local ammonia escape caused by local excessive ammonia supply in terms of spatial distribution.

[0164] S4: Input the distributed ammonia injection feedforward compensation sequence into the SCR feedforward feedback composite controller, generate the final ammonia injection control command based on the distributed ammonia injection feedforward compensation sequence, and drive the ammonia injection actuator to complete online adjustment;

[0165] Further, S4 includes the following steps:

[0166] S41: Construct an ammonia injection feedforward compensation queue based on the allocation-type ammonia injection feedforward compensation sequence; the ammonia injection feedforward compensation queue is a first-in-first-out queue data structure, and the queue elements are a vector of ammonia injection setpoints arranged according to the ammonia injection loop number. The queue length of the ammonia injection feedforward compensation queue is determined by the preset prediction steps to ensure that the ammonia injection feedforward compensation queue can cover the comprehensive delay window from ammonia injection to the outlet measuring point.

[0167] S42: Extract the first element of the current queue based on the ammonia injection feedforward compensation queue and perform benchmark superposition processing to obtain the current feedforward ammonia injection setpoint vector; the benchmark superposition processing is to map the theoretical ammonia injection quantity obtained in S31 to the theoretical allocated ammonia injection vector according to the allocation calculation processing in S37, and perform algebraic synthesis of the theoretical allocated ammonia injection vector with the first element of the ammonia injection feedforward compensation queue to obtain the current feedforward ammonia injection setpoint vector for the current control cycle.

[0168] S43: Perform final fusion processing based on the current feedforward ammonia injection setpoint vector and the feedback correction ammonia injection quantity to obtain the final ammonia injection control command; the final fusion processing is to decompose the feedback correction ammonia injection quantity into a loop correction vector according to the ammonia injection loop number and superimpose it onto the current feedforward ammonia injection setpoint vector, and at the same time apply output limit and rate of change restriction to the final ammonia injection control command. The output limit is obtained by converting the maximum opening degree and minimum opening degree of the ammonia injection valve, and the rate of change restriction is obtained by converting the maximum allowable opening change rate of the ammonia injection actuator, so as to ensure that the final ammonia injection control command can be stably tracked by the actuator.

[0169] S44: Based on the final ammonia injection control command, the ammonia injection actuator drive processing is executed to obtain the actuator drive signal; the ammonia injection actuator drive processing includes valve opening control and ammonia injection flow closed-loop correction. Valve opening control converts the final ammonia injection control command into a valve opening setting. Ammonia injection flow closed-loop correction performs a small correction on the valve opening setting based on layered ammonia injection flow feedback to compensate for valve characteristic nonlinearity and ammonia supply pressure fluctuations. See also Figure 6This is a schematic diagram of the ammonia injection actuator provided in this application embodiment. As shown in the figure, the diagram uses a classic control system block diagram to illustrate the complete signal flow from the final ammonia injection control command to the ammonia injection valve action. After the final ammonia injection control command enters the system, it splits into two paths at the branch point: one path directly enters the valve opening controller, and after conversion, generates an opening reference signal; the other path performs a difference calculation with the stratified ammonia injection flow feedback signal at the comparison node, and the resulting deviation signal is processed by the flow closed-loop correction stage to generate a correction amount. The two signals are superimposed at the summation node to form the final valve opening setting, driving the actuator marked "M" to actuate the ammonia injection valve and complete the precise adjustment of the ammonia flow rate. The feedback loop at the bottom of the figure represents the closed-loop path of the stratified ammonia injection flow signal returning from the valve output to the comparison node. In actual ammonia injection systems, the valve characteristic curve is nonlinear, and the pressure of the ammonia supply network will fluctuate due to the operating status of the upstream ammonia station. These factors will cause a deviation between the valve opening and the actual ammonia injection flow rate. The flow closed-loop correction process compares the command flow rate with the feedback flow rate in real time to make small online corrections to the valve opening, compensating for the effects of valve characteristic nonlinearity and pressure fluctuations, and ensuring that the intended setting of the ammonia injection feedforward compensation sequence can be accurately tracked by the actuator.

[0170] S45: Based on the actuator drive signal and the real-time running sample vector, a rolling update process is performed on the ammonia injection feedforward compensation queue to obtain the updated ammonia injection feedforward compensation queue. The rolling update process involves dequeuing the first element of the ammonia injection feedforward compensation queue in each control cycle and enqueuing the last element in the allocated ammonia injection feedforward compensation sequence that is aligned with the latest short-time temperature prediction sequence, thereby ensuring that the updated ammonia injection feedforward compensation queue maintains time consistency with the rolling ammonia storage dynamic response diagram. (See also...) Figure 7This is a schematic diagram of the ammonia injection feedforward compensation queue provided in this application embodiment. As shown in the figure, the diagram illustrates the first-in-first-out (FIFO) data structure and rolling update mechanism of the ammonia injection feedforward compensation queue. The main body of the queue consists of multiple setpoint vector elements, from left to right: the setpoint vector at the current time T (filled in blue, indicating the head of the queue), the setpoint vector for the predicted time T+1, the intermediate predicted time indicated by ellipses, and the setpoint vector for the predicted time T+N. The dashed box around the queue is labeled "Ammonia Injection Feedforward Compensation Queue (FIFO Structure)," where FIFO is the abbreviation for First-In-First-Out. The bottom of the diagram shows the dequeue operation: the head element is extracted along the arrow direction and used for instruction synthesis in the current control cycle. The right side of the diagram shows the enqueue operation: a new setpoint vector (predicted time T+N+1), marked with a dashed border and green fill, is added to the tail of the queue along the arrow direction, realizing rolling updates. The time look-ahead direction label above illustrates the temporal arrangement of queue elements from the present to the future. In ammonia injection control scenarios for automatic power generation control and peak shaving, there is a significant overall delay from the ammonia injection valve action to the response at the outlet measuring point. By organizing the multi-step look-ahead ammonia injection setpoints in a queue, the control system can adjust the ammonia supply in advance before the outlet measuring point reflects the delay, giving the ammonia injection setpoints a continuous look-ahead structure on the time axis, effectively reducing the impact of temperature gradient-dominated systematic mismatch on control quality.

[0171] S46: Based on the rolling ammonia storage dynamic response diagram, calculate the predicted outlet nitrogen oxide concentration and the predicted ammonia slip concentration to obtain the predicted emission sequence; the method for calculating the predicted outlet nitrogen oxide concentration and the predicted ammonia slip concentration is as follows: convert the final ammonia injection control command into the inlet ammonia supply equivalent and synthesize it with the equivalent stored ammonia flow sequence at the outlet equivalent time index to obtain the equivalent reaction ammonia equivalent sequence, then combine the inlet nitrogen oxide concentration and flue gas flow to calculate the predicted value of the future outlet nitrogen oxide concentration, and map the part exceeding the ammonia equivalent demand to the predicted ammonia slip concentration to form a predicted emission sequence with the same dimension as the actual measurement point.

[0172] S47: Based on the predicted emission sequence and the measured outlet nitrogen oxide concentration and measured ammonia slip concentration in the real-time operating sample vector, a deviation calculation process is performed to obtain a predicted deviation sequence; the predicted deviation sequence is a discrete sequence of predicted value minus measured value, and is stored separately according to the outlet nitrogen oxide concentration deviation and ammonia slip concentration deviation to support parameter write-back for different deviation sources.

[0173] S48: Calculate the prediction deviation threshold based on the prediction deviation sequence and perform over-limit discrimination processing to obtain an online adaptive write-back request; the prediction deviation threshold is calculated by statistically analyzing the robustness scale of the prediction deviation sequence within the time window corresponding to the steady-state operating condition segment set and generating a threshold, the robustness scale being converted from the quantile interval of the prediction deviation sequence; the over-limit discrimination processing generates an online adaptive write-back request when the prediction deviation sequence exceeds the prediction deviation threshold within a continuous discrimination window, the length of the continuous discrimination window being jointly determined by the response time of the outlet measuring point analyzer and the combined delay from ammonia injection to the outlet, to avoid short-term measurement jitter triggering invalid write-back.

[0174] S49: Based on the online adaptive write-back request and index tree update input queue, perform incremental write-back processing to obtain the updated catalyst temperature storage index tree;

[0175] Furthermore, the incremental write-back process includes the following steps:

[0176] S491: Based on the online adaptive write-back request, extract the write-back sample subset from the index tree update input queue to obtain the parameter write-back sample set; the parameter write-back sample set is a subset of the working condition feature sample set that intersects with the time interval of the limit discrimination window, and carries the deviation type identifier that triggers the write-back, so as to distinguish between ammonia deficiency write-back mainly based on the deviation of outlet nitrogen oxide concentration and ammonia excess write-back mainly based on the deviation of ammonia escape concentration.

[0177] S492: Based on the parameter write-back sample set and the catalyst temperature storage index tree, perform leaf node positioning processing to obtain a set of write-back leaf nodes; the leaf node positioning processing is to retrieve the corresponding leaf nodes in the catalyst temperature storage index tree according to the temperature range and flue gas space velocity range in the parameter write-back sample set, and aggregate them according to the catalyst layer number and the catalyst spatial unit index key to form a set of write-back leaf nodes.

[0178] S493: Based on the set of leaf nodes written back and the set of parameter write-back samples, perform parameter recursive update processing to obtain the updated parameter record set; the parameter recursive update processing prioritizes updating the storage capacity temperature slope parameter and the dynamic response time parameter, and performs small step correction on the ammonia saturation storage capacity parameter and the balanced ammonia coverage parameter. The update adopts recursive least squares identification processing with forgetting factor and uses the confidence parameter as the suppression factor of update gain to avoid low confidence nodes being excessively pulled by a single abnormal sample.

[0179] S494: Perform node write-back processing on the catalyst temperature storage index tree based on the updated parameter record set to obtain the updated catalyst temperature storage index tree; the node write-back processing writes the updated parameter record set into the corresponding leaf node and synchronously updates the confidence parameter and deactivation coefficient, so that the updated catalyst temperature storage index tree directly reflects the latest device characteristics during the subsequent S2 retrieval.

[0180] S410: Based on the updated catalyst temperature storage index tree and the real-time running sample vector, trigger the reconstruction of the rolling ammonia storage dynamic response map for the next control cycle to obtain the rolling ammonia storage dynamic response map for the next cycle; the triggering of the reconstruction of the rolling ammonia storage dynamic response map for the next control cycle is to use the updated catalyst temperature storage index tree as the retrieval data source of S25, so as to regenerate the ammonia adsorption and desorption flux distribution and update the rolling ammonia storage dynamic response map in the next control cycle using updated parameters.

[0181] Specifically, S4 embeds the distributed ammonia injection feedforward compensation sequence into the SCR feedforward feedback composite controller in the form of an ammonia injection feedforward compensation queue. This gives the ammonia injection setpoint a continuous look-ahead structure on the time axis, allowing for early adjustment of ammonia supply before the outlet measuring point reflects the change, reducing the impact of temperature gradient-dominated systemic mismatch on control quality. By predicting emission sequences and deviation sequences, the control system not only corrects current deviations but also determines whether the deviation originates from catalyst temperature storage index tree parameter drift. It then uses online adaptive write-back requests to drive incremental correction of the updated catalyst temperature storage index tree, ensuring the model remains consistent with catalyst deactivation and changes in operating conditions. Without online adaptive write-back requests and updated catalyst temperature storage index trees, ammonia injection feedforward compensation would remain fixed on historical identification results, and compensation errors during rapid temperature changes would gradually accumulate and translate into ammonia escape or excessive outlet nitrogen oxide emissions. The output limit and rate of change constraint of the ammonia injection feedforward compensation queue ensure the compensation is feasible in engineering practice, avoiding frequent valve actuations and flow oscillations caused by sudden changes in predicted compensation. This maintains actuator lifespan and control stability while meeting emission constraints.

[0182] In summary, this method parameterizes and makes searchable the catalyst ammonia storage capacity and its sensitivity to temperature changes using a catalyst temperature storage index tree. Then, it uses a rolling ammonia storage dynamic response map to uniformly encode the adsorption-desorption flux driven by the temperature change rate and the transmission delay along the path. Thus, at the control level, the storage release is equivalent to an equivalent stored ammonia flow sequence that can be directly superimposed on the theoretical ammonia injection amount, ultimately forming a queued and distributed distribution-type ammonia injection feedforward compensation sequence. Compared to conventional feedforward feedback control that relies solely on inlet and outlet nitrogen oxide concentrations and flue gas flow, this scheme explicitly models and compensates for the storage balance migration caused by rapid temperature changes as an independent disturbance source. This reduces the risk of excessive ammonia due to desorption superposition during the temperature rise phase before the measurement point lag occurs, and compensates for the risk of ammonia deficiency due to adsorption occupation during the temperature drop phase before the outlet nitrogen oxide concentration surges. Simultaneously, the online adaptive write-back mechanism continuously corrects the catalyst temperature storage index tree and the rolling ammonia storage dynamic response map as the unit operates over a long period, avoiding misjudging inevitable catalyst deactivation and operating condition migration as control problems requiring increased feedback gain. This suppresses oscillations and overshoot from the control mechanism perspective, demonstrating unique adaptability and irreplaceability for large-scale load changes and significant temperature transitions in automatic power generation control peak shaving.

[0183] For example, after a certain automatic power generation control peak shaving command is triggered, the control system first reads the real-time operating sample vector by S21 and obtains the real-time temperature change rate vector by S22, and then generates a short-time temperature prediction sequence by S23. Subsequently, S25 retrieves the local ammonia storage parameter vector set from the catalyst temperature storage index tree using the short-time temperature prediction sequence and the space velocity prediction sequence, and obtains the ammonia adsorption and desorption flux distribution by S29 and S210. Based on this, S212 constructs a rolling ammonia storage dynamic response diagram and S33 forms an equivalent stored ammonia flow sequence. S34 outputs an ammonia injection feedforward compensation sequence and S37 outputs a distribution-type ammonia injection feedforward compensation sequence. Finally, S41 to S44 issue the distribution-type ammonia injection feedforward compensation sequence as the final ammonia injection control command in the form of an ammonia injection feedforward compensation queue and drive the actuator drive signal to complete the ammonia injection regulation. At the same time, S46 to S49 trigger the updated catalyst temperature storage index tree to be written back and updated according to the prediction deviation sequence, so that the rolling ammonia storage dynamic response diagram of the next control cycle is reconstructed using the updated catalyst temperature storage index tree.

[0184] Example 2:

[0185] This embodiment, based on Embodiment 1, provides a feedforward feedback composite control system for ammonia injection quantity in an SCR denitrification system, such as... Figure 8 As shown, it includes:

[0186] Data acquisition and index tree construction module: used to collect historical and real-time operating data of SCR system and form a set of operating condition feature samples, and build a catalyst temperature storage index tree based on the set of operating condition feature samples;

[0187] Ammonia storage dynamic response map generation module: used to infer the ammonia adsorption and desorption flux distribution based on the catalyst temperature storage index tree and real-time temperature change rate, and obtain the ammonia storage dynamic response map;

[0188] Ammonia injection feedforward compensation sequence generation module: used to calculate the equivalent stored ammonia flow rate based on the ammonia storage dynamic response diagram and superimpose it with the theoretical ammonia injection rate to generate a distribution type ammonia injection feedforward compensation sequence;

[0189] Ammonia injection control and execution module: It is used to input the ammonia injection feedforward compensation sequence into the SCR feedforward feedback composite controller, generate the final ammonia injection control command based on the ammonia injection feedforward compensation sequence, and drive the ammonia injection actuator to complete online adjustment.

Claims

1. A method for feedforward and feedback composite control of ammonia injection quantity in an SCR denitrification system, characterized in that, The method includes: S1: Collect historical and real-time operating data of the SCR system and form a set of operating condition feature samples. Construct a catalyst temperature storage index tree based on the set of operating condition feature samples. The catalyst temperature storage index tree is a hierarchical tree structure. The root node corresponds to the entire SCR reactor. The first-level nodes correspond to the catalyst layer number. The second-level nodes correspond to the temperature range. The third-level nodes correspond to the flue gas space velocity range. The leaf nodes store tree node records. The tree node records include at least the ammonia saturation storage capacity parameter, the equilibrium ammonia coverage parameter, the storage capacity temperature slope parameter, the dynamic response time parameter, the reliability parameter, and the deactivation coefficient. The code in the catalyst spatial unit index table is used as the index key. S2: Based on the catalyst temperature storage index tree and real-time temperature change rate, the ammonia adsorption and desorption flux distribution is inferred to obtain the ammonia storage dynamic response map; S3: Calculate the equivalent stored ammonia flow rate based on the ammonia storage dynamic response diagram and superimpose it with the theoretical ammonia injection rate to generate a distribution-type ammonia injection feedforward compensation sequence; S4: Input the distributed ammonia injection feedforward compensation sequence into the SCR feedforward feedback composite controller, generate the final ammonia injection control command based on the distributed ammonia injection feedforward compensation sequence, and drive the ammonia injection actuator to complete online adjustment.

2. The method for feedforward and feedback composite control of ammonia injection quantity in an SCR denitrification system according to claim 1, characterized in that, The steps for constructing the catalyst temperature storage index tree include: Based on the signal point list of the SCR system, a unified mapping process is performed on the acquisition channels of the distributed control system, the acquisition channels of the flue gas online monitoring system, and the feedback channels of the ammonia injection actuator to obtain a multi-source signal mapping table. Based on the multi-source signal mapping table, historical and real-time acquired data are synchronously processed and completed to obtain a complete running sample matrix; Anomaly removal is performed based on the complete running sample matrix to obtain a cleaned running sample matrix; Based on the cleaning operation sample matrix, the working condition segmentation process is performed to obtain the working condition feature sample set; Based on the working condition feature sample set and the steady-state working condition segment set obtained from the working condition segmentation process, the steady-state ammonia storage parameter identification process is performed to obtain the steady-state ammonia storage parameter table. Based on the working condition feature sample set and the rapid temperature change working condition segment set obtained in the working condition segmentation process, the dynamic parameter identification process of temperature change ammonia storage is performed to obtain the dynamic parameter table of temperature change ammonia storage. A catalyst temperature storage index tree is constructed based on the steady-state ammonia storage parameter table and the variable-temperature ammonia storage dynamic parameter table.

3. The method for feedforward and feedback composite control of ammonia injection quantity in an SCR denitrification system according to claim 2, characterized in that, The steps for synchronizing and completing historical and real-time acquired data include: Based on the multi-source signal mapping table, historical and real-time acquired data are synchronously acquired and archived to obtain a set of multi-source raw operational data. Based on the data acquisition timestamps in the multi-source original running data set, the sampling sequences of each variable are resampled to obtain a unified time axis running sample matrix; Missing point imputation is performed on the running sample matrix based on a unified time axis to obtain a complete running sample matrix; The resampling process employs piecewise linear interpolation with added monotonicity constraints. These constraints limit the increment of the interpolation point to the supremum of the increment of the adjacent original sampling point. The supremum is calculated from the upper quantile of the distribution of the absolute values ​​of the increments of the corresponding variables in the multi-source original running data set.

4. The method for feedforward and feedback composite control of ammonia injection quantity in an SCR denitrification system according to claim 2, characterized in that, The steps for performing condition segmentation processing based on the cleaning operation sample matrix include: Based on the cleaning operation sample matrix, the temperature change rate matrix and the flow rate change rate sequence are calculated to obtain the change rate feature matrix; Based on the rate of change feature matrix, the steady-state discrimination threshold and the rapid temperature change discrimination threshold are calculated to obtain the piecewise threshold parameters; Based on the segmented threshold parameter, the cleaning operation sample matrix is ​​segmented to obtain a set of steady-state operating condition segments and a set of rapid temperature change operating condition segments. Based on the set of steady-state operating condition segments and the set of rapid temperature-changing operating condition segments, sample annotation processing is performed on the cleaning operation sample matrix to obtain the operating condition feature sample set; The set of steady-state operating condition segments is a set of time intervals in which both the rate of temperature change and the rate of flow change do not exceed the steady-state discrimination threshold, and the set of rapid temperature change operating condition segments is a set of time intervals in which the rate of temperature change exceeds the rapid temperature change discrimination threshold and the duration exceeds the minimum segment duration.

5. The method for feedforward and feedback composite control of ammonia injection quantity in an SCR denitrification system according to claim 2, characterized in that, The steps for constructing a catalyst temperature storage index tree based on the steady-state ammonia storage parameter table and the variable-temperature ammonia storage dynamic parameter table include: Based on the structural parameters and catalyst layer arrangement parameters of the SCR reactor, a catalyst spatial unit set is divided, and a catalyst spatial unit index table is obtained. Based on the catalyst spatial unit index table, spatial binding processing is performed on the steady-state ammonia storage parameter table and the variable-temperature ammonia storage dynamic parameter table to obtain a spatially bound ammonia storage parameter set. The confidence parameter and deactivation coefficient are calculated based on the spatially bound ammonia storage parameter set to obtain the confidence deactivation parameter set; Based on the spatially bound ammonia storage parameter set and the confidence deactivation parameter set, a tree node record is constructed to obtain the catalyst temperature storage index tree node set.

6. The method for feedforward and feedback composite control of ammonia injection quantity in an SCR denitrification system according to claim 1, characterized in that, The step of inferring the ammonia adsorption and desorption flux distribution based on the catalyst temperature storage index tree and real-time temperature change rate to obtain the ammonia storage dynamic response map includes: Based on the unified time axis running sample matrix, the running sample vector at the current moment is read from the real-time acquisition window to obtain the real-time running sample vector; Differential smoothing is performed on the historical samples corresponding to the preset number of review steps in the real-time running sample vector and the unified time axis running sample matrix to obtain the real-time temperature change rate vector and the real-time flow change rate value. Short-time temperature prediction processing is performed based on the real-time temperature change rate vector and the real-time running sample vector to obtain a short-time temperature prediction sequence. The flue gas space velocity prediction sequence is calculated based on the short-time temperature prediction sequence and the real-time running sample vector to obtain the space velocity prediction sequence. Based on the short-time temperature prediction sequence, space velocity prediction sequence and catalyst spatial unit index table, an index retrieval process is performed to obtain a set of local ammonia storage parameter vectors. Based on the local ammonia storage parameter vector set and the short-time temperature prediction sequence, the equilibrium storage quantity update process is performed to obtain the target equilibrium ammonia storage quantity sequence. Based on the target balanced ammonia storage sequence and the balanced ammonia coverage parameter, the initial ammonia coverage state quantity estimation process is performed to obtain the initial ammonia coverage state quantity set; Based on the initial set of ammonia coverage state variables, the target equilibrium ammonia storage sequence, and the dynamic response time parameter, dynamic convergence calculation is performed to obtain the predicted sequence of ammonia coverage state variables. Based on the ammonia coverage state quantity prediction sequence and the target equilibrium ammonia storage quantity sequence, the storage ammonia change quantity is calculated to obtain a set of local storage ammonia change quantity prediction sequences. Based on the local storage set of ammonia change prediction sequences and the sampling period, flux conversion processing is performed to obtain the ammonia adsorption and desorption flux distribution; Based on the ammonia adsorption and desorption flux distribution and the catalyst spatial unit index table, a transport delay mapping process is performed to obtain the unit transport delay table. A dynamic response diagram for ammonia storage is constructed based on the ammonia adsorption and desorption flux distribution, the ammonia coverage state quantity prediction sequence, and the unit transport delay table.

7. The method for feedforward and feedback composite control of ammonia injection quantity in an SCR denitrification system according to claim 6, characterized in that, The ammonia storage dynamic response diagram is a directed graph data structure. The nodes of the ammonia storage dynamic response diagram are jointly encoded by the catalyst spatial unit index key and the prediction time index. The node attributes include at least the predicted temperature value, the predicted space velocity value, the predicted value of the ammonia coverage state quantity, and the ammonia adsorption and desorption flux value. The directed edge of the ammonia storage dynamic response diagram connects the upstream nodes and the downstream nodes along the flue gas flow direction, and the flux influence of the upstream node is mapped to the corresponding prediction time index of the downstream node according to the unit transmission delay table.

8. The method for feedforward and feedback composite control of ammonia injection quantity in an SCR denitrification system according to claim 1, characterized in that, The step of calculating the equivalent stored ammonia flow rate based on the ammonia storage dynamic response diagram and superimposing it with the theoretical ammonia injection rate to generate a distribution-type ammonia injection feedforward compensation sequence includes: Based on the target values ​​of inlet nitrogen oxide concentration and outlet nitrogen oxide concentration in the real-time running sample vector and the current inlet flue gas flow rate of the SCR reactor, a stoichiometric conversion process is performed to obtain the theoretical ammonia injection rate. Based on the outlet nitrogen oxide concentration and ammonia escape concentration in the real-time running sample vector, feedback control calculation is performed to obtain the feedback correction ammonia injection amount; Based on traversing the directed edges of the rolling ammonia storage dynamic response graph and accumulating the flux contribution, the equivalent storage ammonia flow sequence is obtained. A feedforward compensation sequence for ammonia injection is obtained by performing feedforward compensation synthesis based on the equivalent stored ammonia flow rate sequence and the theoretical ammonia injection rate. Based on the real-time temperature change rate vector and the unit load change rate, the feedforward compensation weight coefficient and the feedback correction weight coefficient are calculated to obtain the set of weight coefficients; Based on the ammonia injection feedforward compensation sequence, the feedback correction ammonia injection quantity and weight coefficient set, a composite setpoint fusion process is performed to obtain a composite ammonia injection setpoint sequence. Based on the composite ammonia injection setpoint sequence and the stratified ammonia injection flow constraint table, the allocation calculation process is performed to obtain the allocation-type ammonia injection feedforward compensation sequence. The equivalent storage ammonia flow sequence is obtained by mapping and accumulating the ammonia adsorption and desorption flux values ​​on the directed edge pointing to the outlet direction in the ammonia storage dynamic response diagram according to the unit transmission delay table. The components greater than 0 in the equivalent storage ammonia flow sequence indicate that there is an equivalent ammonia supply from the catalyst storage in the future period, and the components less than 0 indicate that there is an equivalent ammonia deficiency caused by catalyst adsorption in the future period.

9. The method for feedforward and feedback composite control of ammonia injection quantity in an SCR denitrification system according to claim 1, characterized in that, The steps of inputting the distributed ammonia injection feedforward compensation sequence into the SCR feedforward feedback composite controller, generating the final ammonia injection control command based on the distributed ammonia injection feedforward compensation sequence, and driving the ammonia injection actuator to complete online adjustment include: Ammonia injection feedforward compensation queue is constructed based on the allocation-type ammonia injection feedforward compensation sequence; The first element of the current queue is extracted based on the ammonia injection feedforward compensation queue, and benchmark superposition processing is performed to obtain the current feedforward ammonia injection setpoint vector. The final fusion process is performed based on the current feedforward ammonia injection setpoint vector and the feedback correction ammonia injection quantity to obtain the final ammonia injection control command. Based on the final ammonia injection control command, the ammonia injection actuator drive processing is executed to obtain the actuator drive signal; The ammonia injection feedforward compensation queue is updated by performing rolling update processing based on the actuator drive signal and the real-time running sample vector. Based on the dynamic response diagram of rolling ammonia storage, the predicted outlet nitrogen oxide concentration and the predicted ammonia escape concentration are calculated to obtain the predicted emission sequence; Based on the deviation calculation processing performed between the predicted emission sequence and the measured outlet nitrogen oxide concentration and measured ammonia slip concentration in the real-time operating sample vector, the predicted deviation sequence is obtained; The prediction deviation threshold is calculated based on the prediction deviation sequence and the over-limit discrimination processing is performed to obtain the online adaptive write-back request. Incremental write-back processing is performed based on the online adaptive write-back request and index tree update input queue to obtain the updated catalyst temperature storage index tree; Based on the updated catalyst temperature storage index tree and the real-time running sample vector, the rolling ammonia storage dynamic response map for the next control cycle is reconstructed to obtain the rolling ammonia storage dynamic response map for the next cycle.

10. A feedforward feedback composite control system for ammonia injection quantity in an SCR denitrification system, used to implement the feedforward feedback composite control method for ammonia injection quantity in an SCR denitrification system as described in any one of claims 1-9, characterized in that, The system includes: Data acquisition and index tree construction module: used to collect historical and real-time operating data of SCR system and form a set of operating condition feature samples, and build a catalyst temperature storage index tree based on the set of operating condition feature samples; Ammonia storage dynamic response map generation module: used to infer the ammonia adsorption and desorption flux distribution based on the catalyst temperature storage index tree and real-time temperature change rate, and obtain the ammonia storage dynamic response map; Ammonia injection feedforward compensation sequence generation module: used to calculate the equivalent stored ammonia flow rate based on the ammonia storage dynamic response diagram and superimpose it with the theoretical ammonia injection rate to generate a distribution type ammonia injection feedforward compensation sequence; Ammonia injection control and execution module: used to input the distributed ammonia injection feedforward compensation sequence into the SCR feedforward feedback composite controller, generate the final ammonia injection control command based on the distributed ammonia injection feedforward compensation sequence, and drive the ammonia injection actuator to complete online adjustment.

Images