Ultralow-temperature denitration system and method for kiln flue gas

Abstract

The invention relates to the technical field of kiln flue gas denitration, and discloses a kiln flue gas ultralow temperature denitration system and method.The method comprises the steps that S1, flue gas data and equipment data are collected and subjected to time alignment, and a synchronous data sequence is obtained; s2, calculating a derivation quantity and a comprehensive change trend quantity according to the synchronous data sequence, and constructing a process feature sequence; s3, establishing a control prediction model to output a control response result, constructing a reaction characterization quantity according to the process feature sequence, constructing a controllability characterization quantity according to the control response result, and generating a denitration reaction state estimation quantity; and S4, calculating an ammonia spraying control quantity and a temperature adjustment control quantity according to the denitration reaction state estimation quantity, applying a change rate constraint according to the controllability characterization quantity, and issuing and executing the ammonia spraying control quantity and the temperature adjustment control quantity. According to the method, self-adaptive distribution of ammonia spraying and temperature adjustment can be achieved under complex working conditions, adjustment oscillation is reduced, and the continuity and reliability of denitration operation are improved.

Description

Technical Field

[0001] This invention relates to the field of kiln flue gas denitrification technology, specifically to an ultra-low temperature denitrification system and method for kiln flue gas. Background Technology

[0002] In kiln flue gas treatment, denitrification typically involves injecting ammonia water as a reducing agent and completing the reaction within a denitrification reactor. In practice, this is often combined with a flue gas temperature control unit to regulate the temperature conditions entering the reactor. Existing systems generally have a continuous emission monitoring system at the flue gas outlet to collect parameters such as nitrogen oxide concentration, flue gas volumetric flow rate, flue gas temperature, and oxygen content. The controller then adjusts the operation of the ammonia water supply unit and the steam heating unit based on this data to achieve emission compliance.

[0003] However, in actual operation, the kiln's operating conditions fluctuate frequently, and changes in flue gas volume, oxygen content, and temperature can cause rapid drift in reaction conditions. Furthermore, the ammonia injection side and the temperature control side are coupled, making it difficult to distinguish their respective contributions to the change in outlet nitrogen oxide concentration when they operate simultaneously. At the same time, the denitrification process exhibits significant dynamic lag; outlet monitoring values ​​often only show changes some time after the control action occurs. Simply relying on instantaneous outlet values ​​for feedback adjustment can easily lead to continuous, superimposed adjustments before a response appears, resulting in problems such as regulation oscillations, excessive ammonia injection, or high temperature control energy consumption.

[0004] Furthermore, the operating status of the denitrification reactor changes over time. For example, changes in channel resistance can affect airflow distribution and reaction effectiveness. When reactor status changes are superimposed with flue gas operating condition fluctuations, traditional control strategies based on empirical rules or single feedback struggle to simultaneously ensure emission stability and operational economy, easily leading to inconsistent control effects and the need for frequent manual parameter tuning under different operating conditions. Therefore, a control method capable of simultaneously characterizing the reaction status and dynamic controllability is needed to support the coordinated regulation of ammonia injection and temperature control, and improve operational stability. Summary of the Invention

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] A kiln flue gas ultra-low temperature denitrification system, wherein:

[0007] A flue gas inlet, which is connected to the kiln flue, is used to introduce flue gas into the flue gas channel;

[0008] A steam heating unit is provided on the flue gas passage for heating and temperature regulation of the flue gas.

[0009] An ammonia supply unit, wherein the ammonia supply unit is used to provide ammonia reducing agent;

[0010] The injection mixing unit is connected to the ammonia supply unit and the flue gas channel, respectively, and is used to inject ammonia into the flue gas and mix it with the flue gas.

[0011] A denitrification reactor, located downstream of the jet mixing unit, is used to remove nitrogen oxides from the flue gas;

[0012] A flue gas outlet, which is connected downstream of the denitrification reactor, is used to send out the treated flue gas;

[0013] A monitoring unit is installed at the steam heating unit, the denitrification reactor, and the flue gas outlet to collect flue gas data.

[0014] The controller is electrically connected to the monitoring unit, the steam heating unit, and the ammonia supply unit, and generates ammonia injection control commands and temperature adjustment control commands based on flue gas data.

[0015] As a preferred embodiment of the ultra-low temperature denitrification system for kiln flue gas described in this invention, the denitrification reactor sequentially comprises a protective agent bed, a denitrification bed, and a spare bed.

[0016] The protective agent bed is used to protect the flue gas entering the denitrification reactor;

[0017] The denitrification bed is used to remove nitrogen oxides from flue gas;

[0018] The backup bed is used to provide backup nitrogen oxide removal capacity.

[0019] A method for ultra-low temperature denitrification of kiln flue gas includes:

[0020] Step S1: Collect flue gas data and equipment data, and align the flue gas data and equipment data in time to obtain a synchronized data sequence;

[0021] Step S2: Calculate the derived quantity and the comprehensive trend quantity based on the synchronous data sequence, and construct the process feature sequence based on the synchronous data sequence, the derived quantity, and the trend quantity;

[0022] Step S3: Establish a control prediction model. Based on the process characteristic sequence, use the control prediction model to output the control response results. Construct reaction characterization quantities based on the process characteristic sequence. Construct controllability characterization quantities based on the control response results. Merge the reaction characterization quantities and controllability characterization quantities to generate the denitrification reaction state estimation quantity.

[0023] Step S4: Calculate the ammonia injection control quantity and temperature control quantity based on the estimated quantity of denitrification reaction state. At the same time, apply the change rate constraint to the ammonia injection control quantity and temperature control quantity based on the controllability characterization quantity, and issue the ammonia injection control quantity and temperature control quantity for execution.

[0024] As a preferred embodiment of the ultra-low temperature denitrification method for kiln flue gas according to the present invention, the flue gas data includes the inlet flue gas temperature and outlet flue gas temperature of the steam heating unit, the inlet pressure and outlet pressure of the denitrification reactor, the nitrogen oxide concentration, flue gas volume flow rate, flue gas temperature and oxygen content at the flue gas outlet.

[0025] The equipment data includes the opening degree of the regulating valve of the steam heating unit, the ammonia flow rate of the ammonia supply unit, and the opening degree of the ammonia regulating valve.

[0026] As a preferred embodiment of the ultra-low temperature denitrification method for kiln flue gas described in this invention, step S2 specifically comprises:

[0027] The derived quantities include nitrogen oxide emission load, ammonia supply equivalent, reactor pressure difference, and steam heating unit temperature rise. The nitrogen oxide concentration at the flue gas outlet is multiplied by the flue gas volume flow rate, and the units are converted to obtain the nitrogen oxide emission load.

[0028] The ammonia supply equivalent is obtained by multiplying the ammonia flow rate of the ammonia supply unit by the ammonia equivalence coefficient, wherein the ammonia equivalence coefficient is determined based on the ammonia concentration provided by the ammonia supply unit.

[0029] The pressure difference between the inlet and outlet pressures of the denitrification reactor is obtained by subtracting the outlet pressure of the denitrification reactor from the inlet pressure.

[0030] The temperature rise of the steam heating unit is obtained by subtracting the temperature of the inlet flue gas from the outlet flue gas temperature of the steam heating unit.

[0031] The comprehensive trend of change includes the trend of change of nitrogen oxide emission load, the trend of change of ammonia supply equivalent, the trend of change of reactor pressure difference, the trend of change of flue gas temperature at flue gas outlet, and the trend of change of oxygen content.

[0032] Based on the sampling time of the synchronous data sequence, the opening degree of the steam heating unit regulating valve, the opening degree of the ammonia water regulating valve, the concentration and derivative amount of nitrogen oxides at the flue gas outlet, and the comprehensive trend quantity in the synchronous data sequence are spliced ​​together to form process characteristic items. The process characteristic items at each sampling time are arranged in chronological order to obtain the process characteristic sequence.

[0033] As a preferred embodiment of the ultra-low temperature denitrification method for kiln flue gas described in this invention, the control prediction model includes:

[0034] Taking the sampling time t as the current time, the most recent L consecutive sampling times are extracted to form a sliding window, and the derived quantities and comprehensive trend quantities are extracted from the process feature sequence to form the working condition feature sequence.

[0035] Within the sliding window, calculate the changes in nitrogen oxide concentration at the flue gas outlet, the changes in the opening of the ammonia water regulating valve, and the changes in the opening of the steam heating unit regulating valve.

[0036] Based on the correlation between the change in the opening of the ammonia water regulating valve and the change in the opening of the steam heating unit regulating valve, the two changes are decorrelated to obtain the changes on the ammonia injection side and the changes on the temperature regulation side.

[0037] The working condition feature sequence within the sliding window is converged to obtain the working condition converged vector;

[0038] A response kernel dictionary for the ammonia injection side and a response kernel dictionary for the temperature regulation side are pre-set. Each dictionary consists of several candidate response kernels. Based on the operating condition convergence vector, the gating weights are calculated for each candidate response kernel in the dictionary.

[0039] The candidate response kernels are weighted and combined according to the gating weights of the ammonia injection side and the temperature control side respectively to generate the ammonia injection side response kernel and the temperature control side response kernel.

[0040] The predicted contribution of the ammonia injection side is obtained by superimposing the response kernel of the ammonia injection side with the change of the ammonia injection side according to the lag position.

[0041] The predicted contribution of the temperature-controlled side is obtained by superimposing the temperature-controlled side response kernel and the temperature-controlled side change according to the lag position.

[0042] The predicted contribution from the ammonia injection side and the predicted contribution from the temperature regulation side are superimposed to obtain the predicted value of the export increment;

[0043] Based on the ammonia injection side response kernel and the temperature control side response kernel respectively, the hysteresis prediction and the response amplitude prediction are calculated, and the ammonia injection side hysteresis prediction, the ammonia injection side response amplitude prediction, the temperature control side hysteresis prediction, and the temperature control side response amplitude prediction are output as control response results.

[0044] As a preferred embodiment of the ultra-low temperature denitrification method for kiln flue gas according to the present invention, the estimated quantities of the denitrification reaction state include:

[0045] Based on the process characteristic sequence corresponding to sampling time t, the derived quantity, the comprehensive trend quantity, and the nitrogen oxide concentration at the flue gas outlet are spliced ​​together in a pre-set order to obtain the reaction characterization quantity;

[0046] The control response results corresponding to sampling time t are spliced ​​together in a pre-set order to construct the controllability characterization quantity of sampling time t;

[0047] By fusing the reaction characterization quantity and the controllability characterization quantity at sampling time t, the estimated quantity of the denitrification reaction state at sampling time t is obtained.

[0048] As a preferred embodiment of the ultra-low temperature denitrification method for kiln flue gas described in this invention, the ammonia injection control quantity and temperature control quantity include:

[0049] At sampling time t, the nitrogen oxide concentration deviation is calculated based on the estimated amount of denitrification reaction state.

[0050] The nitrogen oxide concentration deviation was scaled and normalized using the predicted response amplitude of the ammonia injection side and the predicted response amplitude of the temperature control side, respectively, to obtain the normalized deviation of the ammonia injection side and the normalized deviation of the temperature control side.

[0051] Calculate the ammonia injection control quantity at sampling time t based on the normalized deviation of the ammonia injection side and the ammonia injection control quantity at the previous sampling time.

[0052] The temperature control quantity at sampling time t is calculated based on the normalized deviation of the temperature control side and the temperature control quantity at the previous sampling time.

[0053] As a preferred embodiment of the ultra-low temperature denitrification method for kiln flue gas described in this invention, the rate of change constraint includes:

[0054] Using the lag prediction quantities of the ammonia injection side and the temperature control side in the controllability characterization quantities, the allowable change rate of the ammonia injection side and the temperature control side are adaptively scaled, respectively. The larger the lag prediction quantity, the smaller the allowable change rate.

[0055] The beneficial effects of this invention are as follows: By constructing a denitrification reaction state estimation quantity, the method of this invention integrates reaction state information and dynamic controllability information into the control decision, so that ammonia injection and temperature adjustment no longer rely solely on the instantaneous monitoring value at the outlet, thereby effectively reducing the risk of over-adjustment caused by process lag; by using the control response result to constrain the rate of change of the control quantity, the control action is made more in line with the actual reaction rhythm, improving the stability of coordinated adjustment; under complex operating conditions, it can achieve adaptive allocation of ammonia injection and temperature adjustment, reduce adjustment oscillation and improve the continuity and reliability of denitrification operation, while also helping to reduce ammonia consumption and improve the overall operating effect. Attached Figure Description

[0056] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0057] Figure 1 This is a schematic diagram of a kiln flue gas ultra-low temperature denitrification system provided in Embodiment 1 of the present invention;

[0058] Figure 2This is an overall flow chart of a method for ultra-low temperature denitrification of kiln flue gas provided in Embodiment 1 of the present invention. Detailed Implementation

[0059] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0060] Example 1, as Figure 1 As shown, this embodiment provides a kiln flue gas ultra-low temperature denitrification system, including,

[0061] The flue gas inlet P1 is connected to the kiln flue. The flue gas enters the flue gas channel through the flue gas inlet and flows along the flue gas channel. Figure 1 The flue gas is represented by a solid line. The flue gas passage is used to transport the flue gas between the steam heating unit, the injection mixing unit, and the denitrification reactor.

[0062] Steam heating unit Z is installed in the flue gas passage and is used to heat and regulate the temperature of the flue gas. Figure 1 The steam is represented by a dashed line. The steam heating unit includes a steam flow detection component Z1, a steam regulating valve Z2, and a steam heater Z3. Steam enters the steam heater through the steam regulating valve and exchanges heat with the flue gas in the flue gas passage, causing the flue gas temperature to change according to the temperature control command. The steam flow detection component is used to detect the steam flow.

[0063] Ammonia supply unit A is used to supply ammonia reducing agent. Figure 1 The ammonia water is represented by a dotted line. The ammonia water supply unit includes an ammonia water storage tank A1, an ammonia water pump A2, an ammonia water regulating valve A3, and an ammonia water flow detection component A4. After being pressurized by the ammonia water pump, the ammonia water enters the injection mixing unit through the ammonia water regulating valve. The ammonia water flow rate is output by the ammonia water flow detection component.

[0064] The injection mixing unit Q is connected to both the ammonia supply unit and the flue gas channel. It is used to inject ammonia into the flue gas and mix it with the flue gas. The injection mixing unit injects ammonia from the ammonia supply unit into the flue gas channel to form an injection jet, which creates disturbance and homogenization within the flue gas channel, so that the ammonia and flue gas reach a predetermined mixing state before entering the denitrification reactor.

[0065] Denitrification reactor B, located downstream of the jet mixing unit, is used to remove nitrogen oxides from flue gas. The reactor comprises two protective agent beds (B1), a denitrification bed (B2), and a backup bed (B3). The protective agent beds provide protective treatment for the flue gas entering the reactor, the denitrification beds remove nitrogen oxides, and the backup bed provides backup nitrogen oxide removal capacity. The reactor has an inlet and an outlet, with monitoring units at both to obtain the inlet and outlet pressures.

[0066] Flue gas outlet P2 is connected to the downstream of the denitrification reactor and is used to send out the treated flue gas. A monitoring unit for emission monitoring is installed at the flue gas outlet to collect the nitrogen oxide concentration, flue gas volumetric flow rate, flue gas temperature, and oxygen content at the flue gas outlet.

[0067] The monitoring unit is installed at the steam heating unit, the denitrification reactor, and the flue gas outlet to collect flue gas data.

[0068] The controller is electrically connected to the monitoring unit, steam heating unit, and ammonia supply unit, and is used to generate ammonia injection control commands and temperature adjustment control commands based on flue gas data. The controller includes a data processing module, a feature construction module, a control prediction module, and a collaborative control module, specifically:

[0069] The data processing module collects flue gas data and equipment data, aligns the flue gas data and equipment data in time, and obtains a synchronized data sequence.

[0070] The feature construction module calculates the derived quantity based on the synchronized data sequence, and constructs the process feature sequence based on the synchronized data sequence and the derived quantity;

[0071] The control prediction module establishes a control prediction model, outputs control response results based on the process characteristic sequence, constructs reaction characterization quantities based on the process characteristic sequence, constructs controllability characterization quantities based on the control response results, and integrates the reaction characterization quantities and controllability characterization quantities to generate a denitrification reaction state estimation quantity.

[0072] The collaborative control module calculates the ammonia injection control quantity and temperature control quantity based on the estimated quantity of the denitrification reaction state. At the same time, it imposes constraints on the rate of change of the ammonia injection control quantity and temperature control quantity based on the controllability characterization quantity, and then issues the ammonia injection control quantity and temperature control quantity for execution.

[0073] Example 2, refer to Figure 2 As one embodiment of the present invention, a method for ultra-low temperature denitrification of kiln flue gas is provided, comprising:

[0074] Step S1: Collect flue gas data and equipment data, and align the flue gas data and equipment data in time to obtain a synchronized data sequence.

[0075] Collect flue gas data and equipment data. The flue gas data includes the inlet and outlet flue gas temperatures of the steam heating unit, the inlet and outlet pressures of the denitrification reactor, and the nitrogen oxide concentration, flue gas volume flow rate, flue gas temperature, and oxygen content at the flue gas outlet.

[0076] The equipment data includes the opening degree of the regulating valve of the steam heating unit, the ammonia flow rate of the ammonia supply unit, and the opening degree of the ammonia regulating valve.

[0077] Furthermore, flue gas data and equipment data come from different collection locations and different collection devices, and the sampling mechanisms of each data item are independent of each other. Therefore, the sampling time and sampling period of each data item are usually inconsistent. Directly combining data items from different sources at the same calculation time can easily lead to the problem that "data at the same time does not correspond to the same actual operating condition". Therefore, it is necessary to time-align flue gas data and equipment data so that each data item at the same sampling time corresponds to the same operating condition as much as possible, thereby forming a synchronous record of all data on the same time axis.

[0078] When aligning flue gas data with equipment data in time, a target time series corresponding to a unified sampling period is established. Each flue gas and equipment data point is then mapped to the target time series according to its respective sampling time, forming a synchronized record of all data points on the same time axis. Missing and abnormal records that cannot be mapped to the target time series are removed. After time alignment, a synchronized data sequence is output.

[0079] Step S2: Calculate the derived quantity and the comprehensive trend quantity based on the synchronous data sequence, and construct the process feature sequence based on the synchronous data sequence, the derived quantity and the trend quantity.

[0080] Derivatives are calculated based on the synchronous data sequence. These derived quantities include nitrogen oxide emission load, ammonia supply equivalent, reactor pressure differential, and steam heating unit temperature rise. Specifically, the nitrogen oxide concentration at the flue gas outlet is multiplied by the flue gas volumetric flow rate, and the units are converted to obtain the nitrogen oxide emission load. The nitrogen oxide emission load represents the total amount of nitrogen oxides emitted with the flue gas per unit time, thus unifying "concentration" and "flow rate" into a load quantity representation that can be used for control decisions.

[0081] The ammonia supply equivalent is obtained by multiplying the ammonia flow rate of the ammonia supply unit by the ammonia equivalence coefficient. The ammonia supply equivalent represents the effective reducing agent supply level entering the injection mixing unit per unit time, thereby making the execution intensity of the ammonia injection side comparable.

[0082] It should be noted that the ammonia equivalence coefficient is determined based on the ammonia concentration provided by the ammonia supply unit. When the ammonia flow rate of the ammonia supply unit is the volumetric flow rate, the ammonia equivalence coefficient is taken as the mass of pure ammonia contained in a unit volume of ammonia. The calculation method is as follows: first, convert the ammonia concentration to a mass fraction (when the concentration is a percentage, divide the percentage by 100 to get the mass fraction), then obtain the ammonia density based on the ammonia concentration, and finally multiply the ammonia density by the mass fraction of the ammonia concentration to obtain the ammonia equivalence coefficient.

[0083] The reactor pressure difference is obtained by subtracting the outlet pressure of the denitrification reactor from the inlet pressure. The reactor pressure difference represents the channel resistance and operating load changes of the denitrification reactor. In the subsequent calculation of control quantities, the impact of reactor state changes on the control effect is identified.

[0084] The temperature rise of the steam heating unit is obtained by subtracting the temperature of the inlet flue gas from the outlet flue gas temperature of the steam heating unit. The temperature rise of the steam heating unit represents the actual intensity of the steam heating unit's effect on flue gas temperature regulation, providing an indication of the effect of temperature regulation on the subsequent calculation of control quantities.

[0085] Furthermore, to characterize the dynamic changes in kiln flue gas and equipment operation over time, a comprehensive trend quantity is calculated. This comprehensive trend quantity includes the trends in nitrogen oxide emission load, ammonia supply equivalent, reactor differential pressure, flue gas temperature at the outlet, and oxygen content. Using the sampling time of the synchronous data sequence as a reference, let two adjacent sampling times be the (t-1)th and tth sampling times, respectively, and the corresponding sampling time interval be the sampling period. For any physical quantity x (nitrogen oxide emission load, ammonia supply equivalent, reactor differential pressure, steam heating unit temperature rise, flue gas temperature, or oxygen content) whose trend quantity is to be calculated, its trend quantity at the tth sampling time is calculated using the following formula: ;

[0086] Among them, H x This represents the trend of change in x; x t This represents the value of x at sampling time t; x t-1 This represents the value of x at sampling time t-1; T represents the sampling period.

[0087] The trend quantity represents the direction and rate of increase or decrease of the quantity corresponding to adjacent sampling times, so that the subsequent control prediction model includes both the current state and the trend of state change in the input, thereby improving the ability to characterize the response characteristics of control actions and reducing the risk of misjudgment caused by relying solely on instantaneous values.

[0088] When constructing the process feature sequence, the sampling time of the synchronous data sequence is used as the reference. The derived quantity and the comprehensive change trend quantity are spliced ​​with the steam heating unit regulating valve opening, ammonia water regulating valve opening and nitrogen oxide concentration at the flue gas outlet in the synchronous data sequence to form the process feature sequence.

[0089] Within the same sampling time, the derived quantity, the comprehensive trend quantity, the opening of the steam heating unit regulating valve, the opening of the ammonia water regulating valve, and the nitrogen oxide concentration at the flue gas outlet are spliced ​​together in a pre-set order to form the process feature item corresponding to that sampling time. Among them, splicing means combining the data in fixed positions into the same feature vector, which is used to completely represent the reaction state, control execution state and its changing characteristics at that sampling time.

[0090] The pre-set sequence is preferably as follows: steam heating unit regulating valve opening, ammonia water regulating valve opening, nitrogen oxide concentration at flue gas outlet, followed by nitrogen oxide emission load, ammonia supply equivalent, reactor pressure difference, steam heating unit temperature rise, followed by nitrogen oxide emission load, ammonia supply equivalent, reactor pressure difference, flue gas temperature, and the trend of oxygen content changes. This pre-set sequence is not fixed; it only needs to be set before constructing the characteristic sequence of the process and remain unchanged in subsequent calculations.

[0091] Repeat the above process for all sampling times in the synchronous data sequence to obtain a set of process feature terms arranged in chronological order; arrange the process feature terms in the order of sampling times to form a process feature sequence, which is used as the input for subsequent control prediction models.

[0092] Step S3: Establish a control prediction model. Based on the process characteristic sequence, use the control prediction model to output the control response result. Construct a reaction characterization quantity based on the process characteristic sequence. Construct a controllability characterization quantity based on the control response result. Merge the reaction characterization quantity and the controllability characterization quantity to generate a denitrification reaction state estimation quantity.

[0093] When the control prediction model performs calculations at sampling time t, it does not directly use single-point instantaneous values ​​to judge the impact of control actions. Instead, it extracts the most recent L consecutive sampling times to form a sliding window, extracts the key quantities reflecting the operating state within the sliding window, and forms a working condition feature sequence that can represent the current operating conditions. Subsequently, it constructs the outlet change quantity and two types of execution change quantities within the same sliding window, and performs decorrelation processing on the two types of execution change quantities to make the change quantity on the ammonia injection side and the change quantity on the temperature control side more statistically separable. Finally, it performs convergence on the working condition feature sequence to obtain the working condition convergence vector, which is used for the subsequent calculation of gating weights, enabling the control prediction model to adaptively select different dynamic response modes as the operating conditions change.

[0094] Specifically, taking the sampling time t as the current time, a sliding window is formed by extracting the most recent L consecutive sampling times. Within the sliding window, derived quantities and comprehensive trend quantities are extracted from the process feature sequence and concatenated to form the operating condition feature sequence. Each derived quantity and each comprehensive trend quantity corresponds to a component of the operating condition feature.

[0095] The operating condition feature sequence is used to characterize the operating conditions and dynamic changes within the sliding window. Derived quantities can unify multiple basic monitoring quantities into a more suitable dimensional form for control decisions. The comprehensive trend quantity reflects the direction and rate of increase or decrease of each quantity within the sliding window, allowing subsequent prediction processes to simultaneously grasp the current level and the trend. The construction of the operating condition feature sequence follows a fixed-order splicing principle, ensuring that the corresponding item at each sampling time in the operating condition feature sequence has a definite physical meaning and a fixed position, facilitating stable reproduction of subsequent gating calculations.

[0096] Within the sliding window, the changes in nitrogen oxide concentration at the flue gas outlet, the changes in the opening of the ammonia water regulating valve, and the changes in the opening of the steam heating unit regulating valve are calculated and expressed as follows: ; ; ;

[0097] Where, Δy t This represents the change in nitrogen oxide concentration at the flue gas outlet; y t This represents the concentration of nitrogen oxides at the flue gas outlet at sampling time t; y t-1 This represents the concentration of nitrogen oxides at the flue gas outlet at sampling time t-1; Δu t This indicates the change in the opening degree of the ammonia water regulating valve; u t This represents the value of the ammonia water regulating valve opening at sampling time t; u t-1 This represents the value of the ammonia water regulating valve opening at sampling time t; Δv t This indicates the change in the opening degree of the regulating valve in the steam heating unit; v t This represents the value of the steam heating unit regulating valve opening at sampling time t; v t-1 This represents the value of the steam heating unit regulating valve opening at sampling time t. The change is used to represent the incremental change between adjacent sampling times, which can transform the control action from the absolute opening level into the action amplitude, thus more directly corresponding to the dynamic impact of the control action on the outlet response.

[0098] Among them, the change in nitrogen oxide concentration at the flue gas outlet directly describes the change in nitrogen oxide concentration at the flue gas outlet and serves as supervisory information in the training and consistency verification of the control prediction model, enabling the model to learn the correspondence between "execution change and outlet change". The changes in the opening of the ammonia water regulating valve and the opening of the steam heating unit regulating valve serve as the execution change inputs for the ammonia injection side and the temperature regulation side, respectively, providing a basis for subsequent dynamic response identification of each channel.

[0099] Based on the correlation between the change in the opening of the ammonia water regulating valve and the change in the opening of the steam heating unit regulating valve, the two changes are decorrelated to obtain the changes on the ammonia injection side and the changes on the temperature regulation side.

[0100] During kiln operation, the ammonia injection side and the temperature control side may be adjusted simultaneously within the same time period, resulting in a high correlation between the two types of execution changes within the sliding window. Directly using these changes for channel-specific dynamic response identification can easily lead to contribution confusion. The purpose of decorrelation processing is to reduce the correlation between the two types of execution changes within the sliding window, making the changes on the ammonia injection side more concentrated in representing the independent action components of the ammonia injection side, and the changes on the temperature control side more concentrated in representing the independent action components of the temperature control side. This provides a more stable and identifiable input basis for subsequent channel-specific response kernel generation. Within the sliding window, Δu... t and Δv t Orthogonalization is represented as: ; ;

[0101] in, This indicates the change on the ammonia injection side; Indicates the calculation of the inner product; This indicates the change in temperature control.

[0102] It should be noted that the orthogonal decoupling of the ammonia injection side and the temperature control side is not used to change the coupling relationship between the ammonia injection side and the temperature control side in the physical system, nor is it used to decouple the physical mechanism of the kiln flue gas denitrification process. This step is only used as a data processing method within the control prediction model to improve the identifiability and stability of the dynamic response identification of each channel. By decorrelating the execution changes within the sliding window, the input collinearity problem caused by the simultaneous operation of multiple execution channels can be mitigated, allowing the subsequently generated ammonia injection side response kernel and temperature control side response kernel to more accurately reflect the dynamic characteristics of their respective execution channels' response to outlet nitrogen oxides. In this way, without adding sensors or changing the execution structure, the control prediction model's ability to distinguish the dynamic response characteristics of the ammonia injection side and the temperature control side is improved, providing a more stable input basis for the subsequent calculation of hysteresis prediction and response amplitude prediction.

[0103] The load condition feature sequences within the sliding window are aggregated to obtain a load condition aggregated vector. The load condition aggregated vector is a fixed-length vector, with its dimension equal to the sum of the number of derived quantities and the number of comprehensive trend quantities contained in the load condition feature sequence at a single sampling time. Before aggregated load condition features, each component of the load condition features at each sampling time within the sliding window is subjected to min-max normalization, expressed as: ;

[0104] Where M represents the normalized operating condition characteristic component; m represents the value of the operating condition characteristic component before normalization; m min The minimum value of the characteristic component of the operating condition; m max This represents the maximum value of the operating condition characteristic component. The minimum and maximum values ​​of the operating condition characteristic components are obtained from historical operating data within the normal operating condition range.

[0105] The normalized operating condition feature sequences are then converged within a sliding window to obtain the operating condition convergence vector, which is represented as follows: ;

[0106] in, The vector represents the convergence vector of operating conditions; L represents the length of the sliding window; l represents the l-th sampling time within the sliding window; s t-l This represents the sequence of operating conditions at sampling time tl. The operating condition convergence vector is used to compress the operating condition information within the sliding window into a fixed-dimensional vector representation, enabling the control prediction model to perform gating weight calculations based on the unified-dimensional operating condition representation at each sampling time, thereby achieving adaptive switching of the dynamic response mode as the operating conditions change.

[0107] The convergence process is executed in units of sliding windows to ensure that the operating condition convergence vector can reflect the comprehensive operating condition status of the most recent L consecutive sampling times. The operating condition convergence vector, together with the changes in ammonia injection and temperature control, constitutes the key inputs for the subsequent stages of the control prediction model. The operating condition convergence vector is responsible for describing "what operating condition is being observed", while the two types of changes are responsible for describing "how much action has been taken".

[0108] Furthermore, a response kernel dictionary for the ammonia injection side and a response kernel dictionary for the temperature control side are pre-set, with each dictionary consisting of several candidate response kernels. Candidate response kernels are obtained through discrete impulse response identification of historical operating data. Based on the changes in ammonia injection and temperature control sides recorded in the historical operating data, as well as the corresponding changes in nitrogen oxide concentration at the flue gas outlet, the distribution of response coefficients to the outlet changes at different lag positions is calculated. The resulting response coefficient sequence is used as the initial response kernel. Discrete impulse response identification is then performed on both the ammonia injection and temperature control sides to obtain multiple sets of response kernels. These multiple sets of response kernels are then used to form candidate response kernels in the ammonia injection side response kernel dictionary and the temperature control side response kernel dictionary.

[0109] At sampling time t, the control prediction model does not directly give the predicted value of the outlet increment. Instead, it first calculates the gating weight based on the convergence vector of the operating conditions. The candidate response kernels are then weighted and combined using the gating weight to generate the ammonia injection side response kernel and the temperature regulation side response kernel respectively. This allows the dynamic response form to adaptively switch with changes in the operating conditions. In subsequent steps, the prediction contribution and the predicted value of the outlet increment are calculated from the response kernel and the change.

[0110] Specifically, let the maximum lag order be K (K is a positive integer). The maximum lag order is used to limit the coverage of the response kernel in the time dimension, and represents Δu. t (or Δv) t The maximum lag step that may affect the export response; let the dictionary size be J (J is a positive integer), the dictionary size represents the number of candidate response kernels in the response kernel dictionary, used to characterize the number of dynamic response modes available under different operating conditions. Each candidate response kernel is used to describe the change in control execution (Δu) under a certain typical operating condition. t and Δv t The hysteresis distribution characteristics affecting the concentration of nitrogen oxides at the flue gas outlet. The candidate response kernel has a fixed length in the hysteresis position dimension to cover the hysteresis range where the control action may have a significant impact.

[0111] At sampling time t, using the operating condition convergence vector corresponding to that sampling time as input, the corresponding gating weights are calculated for each candidate response kernel in the ammonia injection side response kernel dictionary, and simultaneously for each candidate response kernel in the temperature control side response kernel dictionary, expressed as follows: ; ;

[0112] Where, r i,j In the i-th control channel, the gating score of the j-th candidate response core; i represents the control channel index, i=1 represents the ammonia injection side, i=2 represents the temperature control side; π represents the matrix transpose of the gating parameter vector corresponding to the candidate response kernel.i,j Let represent the gating weights; exp represent the exponential function; and j represent the summation index variable of the candidate response kernels. Each component of the gating parameter vector is initialized to a small random number with a mean of 0, ensuring a nearly uniform initial gating weight distribution across all candidate response kernels. These weights are updated via gradient descent during the training of the prediction model and remain fixed during actual operation.

[0113] Subsequently, the candidate response kernels from the ammonia injection side and the temperature control side are weighted and combined according to their corresponding gating weights to generate the ammonia injection side response kernel and the temperature control side response kernel, respectively, as follows: ;

[0114] Among them, h i,k d represents the response coefficient of the synthesized response kernel at the i-th control channel and hysteresis position k; i,j,k Let represent the kernel coefficient of the j-th candidate response kernel at the hysteresis position k in the i-th control channel. Through a gated weighted combination process, the control prediction model can adaptively switch between multiple candidate dynamic response modes according to the convergence vector of the operating conditions, so that the generated response kernel changes with the operating conditions, rather than being fixed in a single response form.

[0115] It should be noted that the response kernel adopts a discrete lag sequence form to describe the intensity distribution of the influence of control execution changes on the change in nitrogen oxide concentration at the flue gas outlet at different lag positions. Both the ammonia injection side response kernel and the temperature control side response kernel consist of several kernel coefficients arranged according to lag positions. The lag positions extend sequentially from the 0th lag position to the pre-set maximum lag range. Each kernel coefficient corresponds to the contribution weight of the execution change at that lag position to the change in outlet concentration. Therefore, when the response kernel is superimposed on the ammonia injection side change or the temperature control side change according to lag position, the cumulative summation of the contributions at different lag positions is achieved to obtain the predicted contribution of the corresponding channel.

[0116] After obtaining the ammonia injection-side response kernel and the temperature regulation-side response kernel, the corresponding prediction contributions are calculated based on the changes in ammonia injection-side and temperature regulation-side changes obtained within the sliding window. Specifically, the control prediction model superimposes the ammonia injection-side response kernel and the ammonia injection-side change according to the lag position to obtain the ammonia injection-side prediction contribution; simultaneously, it superimposes the temperature regulation-side response kernel and the temperature regulation-side change according to the lag position to obtain the temperature regulation-side prediction contribution. The superposition calculation is used to describe the cumulative impact of historical execution changes on the change in NOx concentration at the flue gas outlet at the current sampling time at different lag positions. The predicted contributions from the ammonia injection-side and the temperature regulation-side are superimposed to obtain the predicted value of the outlet increment, expressed as: ;

[0117] in, represents the predicted export increment, i.e., the predicted change in nitrogen oxide concentration at flue gas outlets; k represents the lag position index; h 1,k This represents the response coefficient of the ammonia injection side response kernel at the hysteresis position k; This represents the value of the change in ammonia injection at sampling time tk; h 2,k This represents the response coefficient of the temperature-controlled side response kernel at the hysteresis position k; This represents the value of the temperature control side change at sampling time tk. The outlet increment prediction value is used to characterize the trend of flue gas outlet nitrogen oxide concentration change predicted by the control prediction model under the current operating conditions and execution changes.

[0118] During the control prediction model establishment phase, historical operating data is acquired and used to train the model. This historical operating data includes historical flue gas data and historical equipment data, derived from actual operating records of the system under different operating conditions. During training, training samples are constructed from the historical operating data using a sliding window approach. Each training sample corresponds to a continuous sampling time window. The process feature sequence within the window is used to generate the operating condition convergence vector. Changes in ammonia injection and temperature control are used as model inputs, and changes in nitrogen oxide concentration at the flue gas outlet are used as supervisory outputs.

[0119] Because ammonia injection and temperature control are highly correlated in real-world operation (e.g., increased load → simultaneous ammonia and temperature increases), the control prediction model has almost no training samples that can distinguish "who caused the change in nitrogen oxide concentration at the flue gas outlet." Therefore, by constructing training samples with orthogonal identification information during the training phase, and without relying on additional sensors, discriminative identification stimuli are actively introduced, enabling the model to obtain clear channel-specific response information during training, thereby improving the discriminability of the dynamic responses of the ammonia injection side and the temperature control side.

[0120] Specifically, to determine whether an identification stimulus needs to be introduced, the changes in ammonia injection and temperature control are extracted from historical operating data using a sliding window. The criteria for this are met when the two sequences within the time window satisfy the following: ;

[0121] in, This indicates the change in ammonia injection side in historical operating data; || represents the change in temperature control in historical operating data; || represents absolute value operation. represents the square root operation; P represents the pre-set correlation threshold. The correlation threshold is determined based on the statistical analysis results of historical operating data. By analyzing the correlation distribution of changes in the ammonia injection side and the temperature control side under different operating conditions, a threshold that can effectively identify highly synchronous change intervals is selected. This ensures the representativeness of the training samples while avoiding the repeated introduction of identification stimuli into time windows that already have the identification conditions.

[0122] Within the required time window, construct the ammonia injection side identification excitation sequence and the temperature control side identification excitation sequence respectively. The ammonia injection side identification excitation sequence and the temperature control side identification excitation sequence can be generated using pre-defined waveform forms, such as amplitude-limited periodic perturbations or pseudo-random sequences, and satisfy the following constraints: ;

[0123] in, This represents the ammonia injection side identification excitation at the t-th sampling time; This represents the temperature-controlled side identification excitation at the t-th sampling time.

[0124] The ammonia injection side identification excitation is superimposed on the ammonia injection side change in the historical operating data according to the sampling time, and the temperature regulation side identification excitation is superimposed on the temperature regulation side change in the historical operating data according to the sampling time. The corresponding flue gas outlet nitrogen oxide concentration change is collected within the same sliding window to construct training samples containing orthogonal identification information.

[0125] During the training of the control prediction model, training samples containing orthogonal identification information are used first. The parameters of the ammonia injection side response kernel dictionary and the temperature regulation side response kernel dictionary are updated respectively. This enables the model to explicitly distinguish the independent contributions of the two execution channels to the outlet response during the training phase and suppress response kernel aliasing caused by the correlation of execution channels.

[0126] Mean squared error (MSE) is used as the loss function to measure the deviation between the predicted outlet increment and the actual change in nitrogen oxide concentration at the flue gas outlet. During training, the control prediction model adjusts the ammonia injection-side response kernel dictionary, the temperature regulation-side response kernel dictionary, and the gating parameters related to the operating condition convergence vector. This gradually reduces the error between the predicted outlet increment calculated based on the generated ammonia injection-side and temperature regulation-side response kernels and the corresponding actual change in nitrogen oxide concentration at the flue gas outlet, given changes in ammonia injection-side and temperature regulation-side parameters. This completes the training of the model parameters. A gradient-based iterative optimization algorithm is preferably used to minimize the MSE loss function, ensuring that the response kernel dictionary and gating parameters are updated progressively in the direction of reducing the loss function. Specifically, during training, training samples constructed using a sliding window are input into the control prediction model in batches. The current loss function value is calculated based on each batch of training samples, and the model parameters are updated until the loss function converges or the preset training rounds are reached.

[0127] After generating the ammonia injection side response kernel and the temperature regulation side response kernel, the control prediction model does not directly use the response kernel as the final output, but further extracts the control response results from the response kernel for subsequent collaborative control calculations.

[0128] For the ammonia injection side response kernel, the ammonia injection side hysteresis prediction is calculated by analyzing the distribution of the response kernel at each hysteresis position. The ammonia injection side hysteresis prediction is used to characterize Δu. t The time delay characteristics that have a major impact on the NOx concentration at the flue gas outlet are analyzed. Similarly, for the temperature-controlled side response kernel, the temperature-controlled side hysteresis prediction is calculated to characterize Δv. t The time delay characteristics that have a major impact on the concentration of nitrogen oxides at the outlet.

[0129] The ammonia injection-side response kernel or temperature control-side response kernel is used to describe the distribution of the impact of control execution changes on the NOx concentration change at the flue gas outlet at different lag positions. In actual operation, the control execution changes do not take effect immediately on the outlet response, but gradually manifest over a certain period of time, and their impact usually reaches its main effect within a certain lag interval. Therefore, the response intensity distribution of the response kernel at each lag position is regarded as the "impact distribution" in the time dimension. By performing a weighted analysis on this distribution, the time position at which the control execution change has a major impact on the outlet response can be determined. A quantity that can characterize the concentrated time position of the main impact is extracted from the response kernel as a lag prediction quantity, which is used to characterize the typical response delay characteristics of the control channel under the current operating conditions.

[0130] Simultaneously, the response intensity of the ammonia injection side response kernel at the lag position is cumulatively calculated to obtain the predicted amplitude of the ammonia injection side response. This predicted amplitude is used to characterize the overall impact of changes in ammonia injection side control execution on the outlet response. Similarly, the predicted amplitude of the temperature control side response is calculated to characterize the overall impact of changes in temperature control execution on the outlet response.

[0131] The response intensity of the ammonia injection-side response kernel or the temperature control-side response kernel at each lag position reflects the contribution of control execution changes to the outlet response under different time delays. By synthesizing these response intensities over the lag range, the overall impact level of the control execution change throughout the entire response process can be obtained. By accumulating the response intensities of the response kernel at each lag position, a response amplitude prediction is obtained, which characterizes the overall effect of the control channel on the change in NOx concentration at the flue gas outlet under the current operating conditions. Since the response amplitude prediction is derived from the overall distribution of the response kernel, rather than the value at a single lag position, it can more stably reflect the comprehensive effect of control execution changes and provide a basis for subsequent control rate constraints.

[0132] The formulas for calculating the lag forecast and the response magnitude forecast are as follows: ; ;

[0133] Among them, D iG represents the hysteresis prediction of the i-th execution channel; || represents taking the absolute value; i This represents the predicted response amplitude of the i-th execution channel.

[0134] The lag prediction obtained in this way does not rely on instantaneous measurement at a single moment, but comprehensively considers the response distribution at multiple lag locations, thereby avoiding lag judgment distortion caused by noise or short-term fluctuations.

[0135] The control prediction model outputs the predicted lag values ​​of the ammonia injection side, the predicted response amplitude of the ammonia injection side, the predicted lag values ​​of the temperature control side, and the predicted response amplitude of the temperature control side as the control response results.

[0136] Furthermore, reaction characterization quantities are constructed based on process characteristic sequences, and controllability characterization quantities are constructed based on control response results. The reaction characterization quantities and controllability characterization quantities are then fused to generate an estimated denitrification reaction state quantity.

[0137] Specifically, at sampling time t, a denitrification reaction state estimation quantity is constructed based on the process characteristic sequence. A comprehensive state quantity that can simultaneously reflect the "reaction state" and "dynamic controllability" supports the subsequent coordinated control of ammonia injection and temperature adjustment, thereby avoiding the risk of lag misjudgment and over-adjustment caused by making control decisions based solely on the instantaneous value of nitrogen oxide concentration at the flue gas outlet.

[0138] Based on the process characteristic sequence corresponding to sampling time t, the derived quantity, the comprehensive trend quantity, and the nitrogen oxide concentration at the flue gas outlet are spliced ​​together in a pre-set order to obtain the reaction characterization quantity.

[0139] Among them, the derived quantity is used to unify the collected raw data into a more suitable dimension and physical meaning expression for control decision-making; the comprehensive trend quantity is used to characterize the direction and rate of increase and decrease of the derived quantity and key flue gas parameters in the time dimension, so that the reaction characterization quantity includes both the current level and the trend of change, thereby improving the sensitivity to rapid changes in operating conditions and reducing the risk of misjudgment caused by relying solely on instantaneous values; the nitrogen oxide concentration at the flue gas outlet, as a direct observation on the emission side, is included in the reaction characterization quantity, so that the reaction characterization quantity includes both supply and demand and equipment status related information, as well as direct result information on the emission side, which facilitates the consistency verification of the effect of control decisions in the future.

[0140] The control response results corresponding to sampling time t are spliced ​​together in a pre-set order to construct the controllability characterization quantity of sampling time t.

[0141] The control response results are output by the control prediction model, including the ammonia injection side hysteresis prediction, the ammonia injection side response amplitude prediction, the temperature control side hysteresis prediction, and the temperature control side response amplitude prediction. The hysteresis prediction is used to characterize the time delay characteristic of the control execution change of the corresponding control channel having a major impact on the NOx concentration at the flue gas outlet. The response amplitude prediction is used to characterize the overall influence intensity of the corresponding control channel on the outlet response. Therefore, the controllability characterization quantities can dynamically reflect whether the ammonia injection side and the temperature control side are "easy to control, how quickly the control effect manifests, and how strong the control effect is" under the current operating conditions.

[0142] It should be noted that the order in which reaction characterization quantities and controllability characterization quantities are concatenated is not a specific fixed order. It is sufficient that an order is determined and remains unchanged during operation. This is because reaction characterization quantities, controllability characterization quantities, and denitrification reaction state estimation quantities are essentially feature vectors formed by concatenating multiple components with defined physical meanings in their positions. As long as the positional relationships of each component within the vector remain consistent during model training, prediction, and control calculations, the control prediction model and subsequent control calculations can correctly identify the physical meaning represented by each component. The order itself does not change the physical meaning of each component or the calculation results.

[0143] Finally, the reaction characterization quantity and controllability characterization quantity at sampling time t are fused to obtain the denitrification reaction state estimate quantity at sampling time t. This allows the denitrification reaction state estimate quantity to carry both reaction state information and controllability information in the same vector. Consequently, when generating ammonia injection control quantity and temperature control quantity, the current load, supply, reactor state, temperature control effect, and the dynamic lag and difference in action intensity of the two control channels can be explicitly considered, thereby improving the stability and consistency of collaborative control decision-making under complex operating conditions.

[0144] Step S4: Calculate the ammonia injection control quantity and temperature control quantity based on the estimated quantity of the denitrification reaction state. At the same time, impose constraints on the rate of change of the ammonia injection control quantity and temperature control quantity based on the controllability characterization quantity, and issue the ammonia injection control quantity and temperature control quantity for execution.

[0145] Furthermore, at sampling time t, the ammonia injection control quantity and temperature control quantity are generated based on the estimated quantity of the denitrification reaction state. The core purpose is to unify the emission-side target constraints and the process-side operating state into the same decision-making link, so that ammonia injection and temperature control form a coordinated regulation, rather than each acting independently and causing mutual cancellation or superposition of excessive amounts.

[0146] Specifically, the nitrogen oxide concentration component at the flue gas outlet is obtained from the estimated denitrification reaction state and compared with the target value of nitrogen oxide concentration to calculate the nitrogen oxide concentration deviation: ;

[0147] Among them, e ty0 represents the nitrogen oxide concentration deviation; y0 represents the nitrogen oxide concentration target value, which refers to the reference value of nitrogen oxide concentration reached at the flue gas outlet, and is a pre-set expected value.

[0148] To ensure that the same deviation produces comparable control intensity on both the ammonia injection side and the temperature control side under different operating conditions, the nitrogen oxide concentration deviation is scaled and normalized using the predicted response amplitudes of the ammonia injection side and the temperature control side, respectively, to obtain the normalized deviation: ;

[0149] Among them; e i,t G represents the normalization bias; i,t ε represents the predicted response amplitude; ε represents a pre-set constant, a positive number that prevents division by zero.

[0150] Based on the ammonia injection side normalization deviation and the ammonia injection control quantity at the previous sampling time, the ammonia injection control quantity at sampling time t is calculated. Similarly, based on the temperature control side normalization deviation and the temperature control quantity at the previous sampling time, the temperature control quantity at sampling time t is calculated, expressed as follows: ; ;

[0151] Among them, U t Indicates the ammonia injection control quantity; U t-1 This indicates the ammonia injection control quantity at the previous moment; w U Indicates the ammonia injection control coefficient; e 1,t Indicates the normalization deviation on the ammonia injection side; V t Indicates the temperature control quantity; V t-1 This indicates the temperature control value at the previous moment; w V Indicates the temperature control coefficient; e 2,t This indicates the normalization deviation on the temperature control side.

[0152] The method of updating the control quantity based on the normalized deviation at the previous sampling time is to enable the ammonia injection control quantity and the temperature control quantity to be adjusted incrementally, thereby avoiding repeated amplification of control actions before the outlet response is reflected due to the dynamic lag in the denitrification process.

[0153] By normalizing the nitrogen oxide concentration deviation according to the predicted response amplitude, the dimensional differences in the strength of different control channels can be eliminated, and the same emission deviation can be converted into a control increment with comparable scale on the ammonia injection side and the temperature control side, thereby ensuring that the control decision can be reasonably allocated according to the actual performance of each channel.

[0154] Based on this, the ammonia injection control coefficient and the temperature control coefficient are used to map the normalized deviation into the control quantity increment, and their physical meaning is the control gain of the control channel. The ammonia injection control coefficient and the temperature control coefficient are pre-set positive parameters, and their values ​​are determined according to the system operation experience or debugging results. They are used to adjust the response speed of the control quantity update, so that the control process gradually approaches the target emission level while ensuring stability.

[0155] Furthermore, constraints are imposed on the rate of change of the ammonia injection control quantity and the temperature control quantity based on the controllability characterization quantities. At sampling time t, the hysteresis prediction quantities of the ammonia injection side and the temperature control side are read from the controllability characterization quantities, and corresponding allowable rates of change are generated for the ammonia injection control quantity and the temperature control quantity, respectively. The allowable rate of change is monotonically inversely related to the hysteresis prediction quantity; the larger the hysteresis prediction quantity, the smaller the allowable rate of change. Subsequently, the controller limits the increment of the ammonia injection control quantity and the increment of the temperature control quantity calculated at sampling time t to the range corresponding to their respective allowable rates of change, so that the changes of the ammonia injection control quantity and the temperature control quantity at sampling time t relative to the previous sampling time do not exceed the allowable range, and then the limited ammonia injection control quantity and the temperature control quantity are issued for execution.

[0156] When the lag prediction increases, the main impact of the control action on the NOx concentration at the flue gas outlet takes longer to manifest. If the control quantity is still allowed to change rapidly, it is easy to continuously add adjustments before the outlet response appears, leading to over-adjustment and oscillation. By reducing the allowable rate of change when the lag prediction is large, the control action can be made smoother and closer to the actual response rhythm of the process, thereby improving the stability of the coordinated control of ammonia injection and temperature regulation.

[0157] After the constraints are met, the controller converts the ammonia injection control quantity into an ammonia injection control command and sends it to the ammonia water supply unit. It also converts the temperature control quantity into a temperature control command and sends it to the steam heating unit. This enables the ammonia water supply unit to adjust the opening of the ammonia water regulating valve according to the ammonia injection control command, and the steam heating unit to adjust the opening of the steam regulating valve according to the temperature control command. This achieves coordinated execution of ammonia injection and temperature control within the controlled boundary.

[0158] If the above functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0159] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0160] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0161] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0162] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A low-temperature denitrification system for kiln flue gas, characterized in that, include: A flue gas inlet, which is connected to the kiln flue, is used to introduce flue gas into the flue gas channel; A steam heating unit is provided on the flue gas passage for heating and temperature regulation of the flue gas. An ammonia supply unit, wherein the ammonia supply unit is used to provide ammonia reducing agent; The injection mixing unit is connected to the ammonia supply unit and the flue gas channel, respectively, and is used to inject ammonia into the flue gas and mix it with the flue gas. A denitrification reactor, located downstream of the jet mixing unit, is used to remove nitrogen oxides from the flue gas; A flue gas outlet, which is connected downstream of the denitrification reactor, is used to send out the treated flue gas; A monitoring unit is installed at the steam heating unit, the denitrification reactor, and the flue gas outlet to collect flue gas data. The controller is electrically connected to the monitoring unit, the steam heating unit, and the ammonia supply unit, and generates ammonia injection control commands and temperature adjustment control commands based on flue gas data.

2. The kiln flue gas ultra-low temperature denitrification system as described in claim 1, characterized in that, The denitrification reactor comprises, in sequence, a protective agent bed, a denitrification bed, and a spare bed; The protective agent bed is used to protect the flue gas entering the denitrification reactor; The denitrification bed is used to remove nitrogen oxides from flue gas; The backup bed is used to provide backup nitrogen oxide removal capacity.

3. A method for ultra-low temperature denitrification of kiln flue gas, characterized in that, include: Step S1: Collect flue gas data and equipment data, and align the flue gas data and equipment data in time to obtain a synchronized data sequence; Step S2: Calculate the derived quantity and the comprehensive trend quantity based on the synchronous data sequence, and construct the process feature sequence based on the synchronous data sequence, the derived quantity, and the trend quantity; Step S3: Establish a control prediction model. Based on the process characteristic sequence, use the control prediction model to output the control response results. Construct reaction characterization quantities based on the process characteristic sequence. Construct controllability characterization quantities based on the control response results. Merge the reaction characterization quantities and controllability characterization quantities to generate the denitrification reaction state estimation quantity. Step S4: Calculate the ammonia injection control quantity and temperature control quantity based on the estimated quantity of denitrification reaction state. At the same time, apply the change rate constraint to the ammonia injection control quantity and temperature control quantity based on the controllability characterization quantity, and issue the ammonia injection control quantity and temperature control quantity for execution.

4. The method for ultra-low temperature denitrification of kiln flue gas as described in claim 3, characterized in that, The flue gas data includes the inlet and outlet flue gas temperatures of the steam heating unit, the inlet and outlet pressures of the denitrification reactor, and the nitrogen oxide concentration, flue gas volumetric flow rate, flue gas temperature, and oxygen content at the flue gas outlet. The equipment data includes the opening degree of the regulating valve of the steam heating unit, the ammonia flow rate of the ammonia supply unit, and the opening degree of the ammonia regulating valve.

5. The method for ultra-low temperature denitrification of kiln flue gas as described in claim 4, characterized in that, Step S2 is as follows: The derived quantities include nitrogen oxide emission load, ammonia supply equivalent, reactor pressure difference, and steam heating unit temperature rise. The nitrogen oxide concentration at the flue gas outlet is multiplied by the flue gas volume flow rate, and the units are converted to obtain the nitrogen oxide emission load. The ammonia supply equivalent is obtained by multiplying the ammonia flow rate of the ammonia supply unit by the ammonia equivalence coefficient, wherein the ammonia equivalence coefficient is determined based on the ammonia concentration provided by the ammonia supply unit. The pressure difference between the inlet and outlet pressures of the denitrification reactor is obtained by subtracting the outlet pressure of the denitrification reactor from the inlet pressure. The temperature rise of the steam heating unit is obtained by subtracting the temperature of the inlet flue gas from the outlet flue gas temperature of the steam heating unit. The comprehensive trend of change includes the trend of change of nitrogen oxide emission load, the trend of change of ammonia supply equivalent, the trend of change of reactor pressure difference, the trend of change of flue gas temperature at flue gas outlet, and the trend of change of oxygen content. Based on the sampling time of the synchronous data sequence, the opening degree of the steam heating unit regulating valve, the opening degree of the ammonia water regulating valve, the concentration and derivative amount of nitrogen oxides at the flue gas outlet, and the comprehensive trend quantity in the synchronous data sequence are spliced ​​together to form process characteristic items. The process characteristic items at each sampling time are arranged in chronological order to obtain the process characteristic sequence.

6. The method for ultra-low temperature denitrification of kiln flue gas as described in claim 5, characterized in that, The control prediction model includes: Taking the sampling time t as the current time, the most recent L consecutive sampling times are extracted to form a sliding window, and the derived quantities and comprehensive trend quantities are extracted from the process feature sequence to form the working condition feature sequence. Within the sliding window, calculate the changes in nitrogen oxide concentration at the flue gas outlet, the changes in the opening of the ammonia water regulating valve, and the changes in the opening of the steam heating unit regulating valve. Based on the correlation between the change in the opening of the ammonia water regulating valve and the change in the opening of the steam heating unit regulating valve, the two changes are decorrelated to obtain the changes on the ammonia injection side and the changes on the temperature regulation side. The working condition feature sequence within the sliding window is converged to obtain the working condition converged vector; A response kernel dictionary for the ammonia injection side and a response kernel dictionary for the temperature regulation side are pre-set. Each dictionary consists of several candidate response kernels. Based on the operating condition convergence vector, the gating weights are calculated for each candidate response kernel in the dictionary. The candidate response kernels are weighted and combined according to the gating weights of the ammonia injection side and the temperature control side respectively to generate the ammonia injection side response kernel and the temperature control side response kernel. The predicted contribution of the ammonia injection side is obtained by superimposing the response kernel of the ammonia injection side with the change of the ammonia injection side according to the lag position. The predicted contribution of the temperature-controlled side is obtained by superimposing the temperature-controlled side response kernel and the temperature-controlled side change according to the lag position. The predicted contribution from the ammonia injection side and the predicted contribution from the temperature regulation side are superimposed to obtain the predicted value of the export increment; Based on the ammonia injection side response kernel and the temperature control side response kernel respectively, the hysteresis prediction and the response amplitude prediction are calculated, and the ammonia injection side hysteresis prediction, the ammonia injection side response amplitude prediction, the temperature control side hysteresis prediction, and the temperature control side response amplitude prediction are output as control response results.

7. The method for ultra-low temperature denitrification of kiln flue gas as described in claim 6, characterized in that, The estimated parameters for the denitrification reaction state include: Based on the process characteristic sequence corresponding to sampling time t, the derived quantity, the comprehensive trend quantity, and the nitrogen oxide concentration at the flue gas outlet are spliced ​​together in a pre-set order to obtain the reaction characterization quantity; The control response results corresponding to sampling time t are spliced ​​together in a pre-set order to construct the controllability characterization quantity at sampling time t; By fusing the reaction characterization quantity and the controllability characterization quantity at sampling time t, the estimated quantity of the denitrification reaction state at sampling time t is obtained.

8. The method for ultra-low temperature denitrification of kiln flue gas as described in claim 7, characterized in that, The ammonia injection control quantity and temperature control quantity include: At sampling time t, the nitrogen oxide concentration deviation is calculated based on the estimated amount of denitrification reaction status. The nitrogen oxide concentration deviation was scaled and normalized using the predicted response amplitude of the ammonia injection side and the predicted response amplitude of the temperature control side, respectively, to obtain the normalized deviation of the ammonia injection side and the normalized deviation of the temperature control side. Calculate the ammonia injection control quantity at sampling time t based on the normalized deviation of the ammonia injection side and the ammonia injection control quantity at the previous sampling time. The temperature control quantity at sampling time t is calculated based on the normalized deviation of the temperature control side and the temperature control quantity at the previous sampling time.

9. The method for ultra-low temperature denitrification of kiln flue gas as described in claim 8, characterized in that, The rate of change constraint includes: Using the lag prediction quantities of the ammonia injection side and the temperature control side in the controllability characterization quantities, the allowable change rate of the ammonia injection side and the temperature control side are adaptively scaled, respectively. The larger the lag prediction quantity, the smaller the allowable change rate.

Images