Medium-temperature selective catalytic reduction denitration intelligent collaborative control system and method
By constructing a dual-channel parallel feedforward architecture for flue gas temperature and ammonia injection, along with a long-term and short-term self-learning mechanism, the control lag and ammonia escape problems of the medium-temperature SCR denitrification system were solved, achieving efficient flue gas treatment and energy consumption optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANXI YUANDIAN DIGITAL INTELLIGENCE TECHNOLOGY CO LTD
- Filing Date
- 2026-05-25
- Publication Date
- 2026-07-07
AI Technical Summary
Existing medium-temperature SCR denitrification systems suffer from severe control response lag, insufficient ammonia slip control precision, high reheat energy consumption, and poor model adaptability, leading to instantaneous NOx exceedances, ammonia slip exceedances, and excessive energy consumption.
By employing a multi-dimensional full-process perception module, a full-process operating condition prediction module, a multi-objective collaborative optimization module, and a collaborative control execution module, a dual-channel parallel feedforward architecture is constructed to synchronously drive flue gas temperature and ammonia injection using LSTM prediction signals. Combined with long and short-term self-learning mechanisms and particle swarm optimization algorithms, collaborative control of flue gas temperature and ammonia injection is achieved.
It achieves advanced response for flue gas temperature regulation and ammonia injection regulation, reduces NOx fluctuations and ammonia escape, reduces hot blast stove gas consumption, and improves denitrification efficiency and catalyst lifespan.
Smart Images
Figure CN122342984A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of flue gas pollution control and intelligent industrial process control technology in the iron and steel metallurgical industry, specifically relating to a medium-temperature selective catalytic reduction denitrification intelligent collaborative control system and method. Background Technology
[0002] Medium-temperature SCR denitrification technology is currently the mainstream technology for denitrification of sintering flue gas in steelmaking. Its reaction activity window is 260-300℃, offering advantages such as high denitrification efficiency and stable operation. Unlike SCR denitrification in coal-fired power plants, sintering flue gas in steelmaking is characterized by lower temperatures, significant temperature fluctuations, high SO2 concentrations, and a narrow catalyst activity window. After semi-dry desulfurization, the flue gas temperature is only 120-140℃, requiring heating to above 260℃ via a gas-fired gas heater (GGH) and hot blast stove to meet the activity temperature requirements of the medium-temperature SCR catalyst. However, existing medium-temperature SCR denitrification systems suffer from the following problems in actual operation:
[0003] Severe control response lag: Existing systems mostly use independent single-loop PID control. When inlet flue gas parameters fluctuate, the control response exhibits a lag of 10-30 seconds, which can easily lead to NO emissions. x Instantaneous exceedance;
[0004] Insufficient precision in ammonia slip control: ammonia injection control is based solely on outlet NO. x The concentration feedback regulation does not take into account the fluctuations in inlet flue gas temperature and SO2 concentration, which can easily lead to excessive ammonia injection, resulting in ammonia escape exceeding the standard and the generation of ammonium bisulfate that clogs the catalyst.
[0005] High energy consumption and large temperature fluctuations in supplementary heating: Existing supplementary heating control relies on feedback adjustment when the temperature drops below the threshold. This not only results in lag in response, leading to drastic fluctuations in flue gas temperature, but also causes a large amount of gas waste due to excessive supplementary heating.
[0006] Poor model adaptability: Most existing prediction models are trained offline in one go, which cannot adapt to the long-term drift of sintering conditions and the slow decay of catalyst activity in real time. They lack global optimization capabilities and are difficult to balance the achievement of denitrification standards in medium-temperature SCR with the economic efficiency of operation. Summary of the Invention
[0007] The purpose of this invention is to provide a medium-temperature selective catalytic reduction denitrification intelligent collaborative control system and method to solve the problems of control lag, excessive ammonia escape, high energy consumption for heat replenishment, and poor model adaptability in the prior art.
[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:
[0009] A medium-temperature selective catalytic reduction denitrification intelligent collaborative control system includes a multi-dimensional full-process perception module, a full-process operating condition prediction module, a multi-objective collaborative optimization module, a collaborative control execution module, and an execution layer module that are connected in sequence via communication.
[0010] The collaborative control execution module includes a flue gas temperature control module and a precision ammonia injection collaborative control module connected in parallel communication.
[0011] The feedback output terminal of the execution layer module is communicatively connected to the feedback input terminal of the multi-dimensional full-process perception module, forming a closed-loop control link for the entire process of medium-temperature SCR denitrification.
[0012] The full-process operating condition prediction module includes a historical operating database unit, a Long Short-Term Memory (LSTM) time series prediction unit, and a feedforward control signal output unit. The historical operating database unit has a built-in offline learning module for training, generating, and updating the prediction model based on historical data of the entire operating condition of the mesophilic SCR denitrification process. The LSTM time series prediction unit is an online learning module used to update model parameters in real time and predict the inlet flue gas temperature and NO₂ of the mesophilic SCR reactor in advance. x The changing trends of key operating parameters such as concentration, CO concentration, and flue gas flow rate, along with the prediction time, enable both flue gas temperature regulation and ammonia injection regulation to have advanced response capabilities.
[0013] The output of the full-process operating condition prediction module is connected to the input of the flue gas temperature control module and the precision ammonia injection coordinated control module, respectively. It is used to synchronously output feedforward prediction signals to the flue gas temperature regulation and ammonia injection regulation before medium temperature SCR denitrification, so that the flue gas temperature regulation and ammonia injection regulation cooperate with each other in time to form a coordinated feedforward control of flue gas temperature and ammonia injection.
[0014] This invention connects the output of the full-process operating condition prediction module to the inputs of the flue gas temperature control module and the precision ammonia injection coordinated control module, respectively. This allows for the synchronous output of feedforward prediction signals to the flue gas temperature regulation and ammonia injection regulation before medium-temperature SCR denitrification, constructing a dual-channel parallel feedforward architecture of "prediction-flue gas temperature-ammonia injection". This architecture enables the flue gas temperature regulation and ammonia injection regulation to coordinate in timing; when NO is predicted... x When the load is about to increase significantly, the flue gas temperature control raises the reaction temperature to the optimal window in advance, and the ammonia injection control matches the ammonia-nitrogen molar ratio in sync, forming a time-series coordinated control, which fundamentally solves the problem of instantaneous exceedance caused by response lag in traditional single-channel feedforward or feedback control.
[0015] The multi-objective collaborative optimization module includes rigidly constrained boundary elements, a multi-objective global optimization element, and an optimal operating parameter solution element. The multi-objective global optimization element employs a particle swarm optimization algorithm, targeting the NO₂ at the outlet of a medium-temperature SCR denitrification system. xThe concentration and ammonia slip concentration meet the ultra-low emission limit as rigid constraints, and the optimization objective is to minimize the overall energy consumption of the denitrification system. The optimal operating parameter solution unit solves the optimal parameter combination of the inlet flue gas temperature setpoint of the medium-temperature SCR reactor, the opening degree of the ammonia injection regulating valve, and the soot blowing sequence of the flue gas-flue gas heat exchanger based on the output results of the particle swarm algorithm.
[0016] The output of the multi-objective collaborative optimization module is communicatively connected to the input of the flue gas temperature control module and the precision ammonia injection collaborative control module, respectively, and is used to output the optimized operating parameter setpoints.
[0017] The flue gas temperature control module is located at the inlet of the medium-temperature SCR reactor and includes a self-heating control sub-logic unit and a supplementary heat switching sub-logic unit. The supplementary heat switching sub-logic unit has a built-in gas consumption prediction model, which is used to receive the flue gas temperature prediction signal output by the full-process operating condition prediction module. When it is predicted that the flue gas temperature at the inlet of the medium-temperature SCR reactor will be lower than the lower limit temperature of the catalyst activity window, the required gas consumption is calculated in real time, and the opening prediction value of the hot air furnace gas regulating valve is output in advance to achieve uninterrupted supplementary heat switching.
[0018] The precise ammonia injection coordinated control module is located in the inlet flue of the mesotemperature SCR reactor and includes a feedforward loop unit, a feedback loop unit, and a correction loop unit. The feedforward loop unit has a built-in long-term and short-term self-learning module, used to simultaneously collect minute-level inlet flue gas parameter mutation data and daily mesotemperature SCR operating trend data, and fuse them to calculate the ammonia injection feedforward value. The feedback loop unit is used to calculate the ammonia injection feedforward value based on the SCR reactor outlet NO... x The concentration fine-tunes the total ammonia injection opening; the correction loop unit is used to limit and correct the ammonia injection rate based on ammonia escape concentration and inlet flue gas temperature data. The long-term and short-term self-learning modules capture instantaneous changes in inlet parameters (such as NO) through two parallel learning channels. x The concentration fluctuates at the second level and the long-term drift of catalyst activity (such as monthly decay trends) are measured. After weighted fusion of the dual-channel signals, an accurate ammonia injection feedforward value is generated, which effectively avoids the under-injection or over-injection problems caused by single time scale learning.
[0019] Furthermore, the lower limit temperature of the catalyst activity window of the supplementary heating switching sub-logic unit is 260℃; the prediction duration of the LSTM time series prediction unit is 5-10 minutes, enabling the flue gas temperature regulation and ammonia injection regulation to have a 5-10 minute lead response capability.
[0020] Furthermore, the short-term and long-term self-learning modules in the feedforward loop unit include a short-term learning channel and a long-term learning channel. The short-term learning channel is used to capture the entry NO in the first cycle. xThe instantaneous fluctuations in concentration and flue gas flow rate are tracked by the long-term learning channel, which is used to track the changing trends of catalyst activity decay and equipment characteristic drift in the second cycle. The first cycle is on the minute level, and the second cycle is on the day level. The outputs of the short-term learning channel and the long-term learning channel are weighted and fused to generate the ammonia injection feedforward value.
[0021] Furthermore, the offline learning module built into the historical operation database unit retrains and updates the basic prediction model based on the newly added historical data of the entire operating condition of medium-temperature SCR denitrification in the first update cycle; the LSTM time series prediction unit uses real-time operation data to fine-tune the model online in the second update cycle; the duration of the first update cycle is longer than the duration of the second update cycle.
[0022] Furthermore, the multi-dimensional full-process sensing module is distributed along the medium-temperature SCR denitrification process route, including a distributed operating condition measurement point subunit, a multi-point flue gas automatic monitoring system (CEMS) component, an equipment status sensor subunit, and a measurement point fault diagnosis and redundancy switching subunit; the execution layer module includes a flue gas temperature control execution unit and an ammonia injection system execution unit.
[0023] Furthermore, the feedforward prediction signal output by the full-process operating condition prediction module simultaneously drives the flue gas temperature control module to perform flue gas temperature feedforward adjustment and the precise ammonia injection collaborative control module to perform ammonia injection feedforward adjustment. The responses of both control channels lead the measured changes in the inlet parameters of the mesophilic SCR reactor. x When the load fluctuates, the flue gas temperature is adjusted in advance to create efficient active temperature conditions for the SCR catalyst, and the ammonia injection is adjusted synchronously to match the ammonia-nitrogen molar ratio, forming a time-sequential synergistic control.
[0024] Furthermore, the system is applied to the medium-temperature SCR denitrification process of sintering flue gas in steelmaking. After semi-dry desulfurization, the flue gas enters the medium-temperature SCR reactor after being reheated by a flue gas-to-flue gas heat exchanger and a hot blast stove. The flue gas temperature control module is used to adjust the opening of the gas regulating valve of the hot blast stove.
[0025] A method for intelligent synergistic control of mesophilic selective catalytic reduction denitrification, based on the above system, includes the following steps:
[0026] S1: Collect the entire process operation data of medium-temperature SCR denitrification through the multi-dimensional full-process sensing module, and complete the data preprocessing and measurement point redundancy switching;
[0027] S2: The full-process operating condition prediction module predicts the parameter change trend of key operating conditions at the inlet of the medium-temperature SCR reactor through an offline + online dual learning mechanism, and simultaneously outputs feedforward control signals to the two channels of the flue gas temperature control module and the precision ammonia injection collaborative control module to realize the collaborative feedforward control of flue gas temperature and ammonia injection.
[0028] S3: The multi-objective collaborative optimization module uses the particle swarm optimization algorithm for global optimization, solves the optimal combination of operating parameters adapted to the medium-temperature SCR denitrification reaction, and sends the optimized flue gas temperature setpoint and ammonia injection opening setpoint to the flue gas temperature control module and the precision ammonia injection collaborative control module, respectively.
[0029] S4: The flue gas temperature control module adjusts the flue gas temperature to the optimal activity temperature window of the medium-temperature SCR catalyst based on the feedforward control signal and the optimized flue gas temperature setpoint.
[0030] Meanwhile, the precision ammonia injection collaborative control module performs long and short-term self-learning feedforward + feedback + correction collaborative ammonia injection control based on the feedforward control signal and the optimized ammonia injection opening setpoint.
[0031] S5: The execution layer module executes control commands, adjusts the operation of the denitrification system and induced draft fan equipment, and sends the operation feedback data back to the multi-dimensional full-process perception module to form a closed-loop control of the entire process.
[0032] Furthermore, the offline + online dual learning mechanism in step S2 is as follows: the LSTM base model is trained offline and the model weights are updated using newly added historical running data in the first update cycle; the model is fine-tuned online using real-time running data in the second update cycle. The first update cycle is longer than the second update cycle, so that the model can continuously adapt to catalyst activity decay and operating condition drift.
[0033] In step S4, the supplementary heating switching sub-logic unit of the flue gas temperature control module receives the flue gas temperature prediction signal. When it is predicted that the flue gas temperature at the reactor inlet will fall below the lower limit temperature of the catalyst activity window within a preset time, the opening prediction value of the hot air furnace gas regulating valve is output in advance, so that the supplementary heating process and the flue gas cooling process are seamlessly connected.
[0034] Compared with the prior art, the beneficial effects of the present invention are:
[0035] 1. This invention differs from existing technologies that only use the prediction signal for ammonia injection feedforward. It constructs a dual-channel parallel feedforward architecture that synchronously drives flue gas temperature control and ammonia injection control using LSTM prediction signals. Both flue gas temperature regulation and ammonia injection regulation possess advanced lead response capabilities, especially in NO... x When the load fluctuates significantly, the flue gas temperature creates an efficient activation temperature for the catalyst in advance, and the ammonia injection is synchronized with the ammonia-nitrogen molar ratio to achieve time-series coordinated control, fundamentally solving the instantaneous exceedance problem caused by the lag in traditional PID control. Experiments show that this architecture, compared with a single-channel feedforward system, results in higher NO output. x The fluctuation range was reduced by more than 40%, and the ammonia slip concentration was reduced by more than 60%.
[0036] 2. This invention deeply couples flue gas temperature prediction feedforward with the supplementary heating switching logic. When the flue gas temperature is predicted to fall below the threshold, the supplementary heating valve is calculated and executed in advance, so that the supplementary heating process is seamlessly connected with the flue gas cooling process. The flue gas temperature fluctuation is controlled within ±3℃, which is nearly 3 times better than the traditional feedback supplementary heating. Combined with the optimal flue gas temperature setpoint output by the PSO algorithm, the SCR reaction is always in the high-efficiency window, while significantly reducing the gas consumption of the hot blast stove, and the overall energy consumption is reduced by about 18%.
[0037] 3. This invention innovatively introduces a long-term and short-term self-learning mechanism in the feedforward loop. Through dual-channel parallel learning, it captures the instantaneous changes in inlet parameters and the long-term drift of catalyst activity, and merges them to generate ammonia injection feedforward values. Then, through the fine adjustment of the feedback loop and the correction loop, the ammonia injection control accuracy is improved by more than 70%, achieving the dual optimization of denitrification efficiency and ammonia slip, effectively avoiding the risk of ammonium bisulfate formation and catalyst blockage, and extending the catalyst service life by more than 20%.
[0038] 4. This invention uses offline model reconstruction and online model fine-tuning with different update cycles, combined with the PSO algorithm, to globally optimize operating parameters such as flue gas temperature, ammonia injection, and soot blowing timing. This enables the system to automatically adapt to long-term changes such as catalyst aging and equipment characteristic drift, maintaining optimal operating status throughout the process, with extremely low maintenance requirements in the later stages. Attached Figure Description
[0039] Figure 1 This is a diagram showing the overall architecture of the medium-temperature SCR denitrification intelligent collaborative control system described in this invention.
[0040] 1. Multi-dimensional full-process perception module; 1.1 Distributed operating condition measurement point sub-unit; 1.2 Multi-point CEMS flue gas monitoring component; 1.3 Equipment status sensor sub-unit; 1.4 Measurement point fault diagnosis and redundancy switching sub-unit;
[0041] 2. Full-process operating condition prediction module; 2.1. Historical operation database unit; 2.2. LSTM time series prediction unit; 2.3. Feedforward control signal output unit;
[0042] 3. Multi-objective collaborative optimization module; 3.1 Rigidly constrained boundary element; 3.2 Multi-objective global optimization element; 3.3 Optimal operating parameter solution element;
[0043] 4. Smoke temperature control module; 4.1. Self-heating control sub-logic unit; 4.2. Replenishment heat switching sub-logic unit;
[0044] 5. Precision ammonia injection coordinated control module; 5.1 Feedforward loop unit; 5.2 Feedback loop unit; 5.3 Correction loop unit;
[0045] 6. Executive layer module; 6.1. Flue gas temperature control execution unit; 6.2. Ammonia injection system execution unit. Specific implementation manner
[0046] To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0047] This embodiment takes the medium-temperature SCR denitration system supporting a 360m 2 sintering machine in a domestic iron and steel plant as the implementation object. The rated flue gas volume of the system is 1100000 Nm 3 / h. The designed reaction temperature window for medium-temperature SCR denitration is 260 - 300 °C, and the ultra-low emission limit is implemented: NO x ≤ 50 mg / Nm3, ammonia slip ≤ 3 ppm. The flue gas temperature after semi-dry desulfurization is only 120 - 140 °C, and it needs to be heated by a GGH and a hot blast stove before entering the medium-temperature SCR reactor.
[0048] After the system is deployed, the historical operation database unit completes the initial model training based on the historical data of the full operating conditions of medium-temperature SCR denitration in the past 6 months. During the operation process, the LSTM time series prediction unit 2.2 collects the operation data at the inlet and outlet of the medium-temperature SCR reactor in real time for online learning, and predicts the key operating parameters 8 minutes in advance. The multi-objective global optimization unit 3.2 uses the particle swarm algorithm for global optimization, the number of iterations is set to 50, the population size is set to 30, and the convergence speed is increased by more than 40% compared with the traditional genetic algorithm.
[0049] To verify the technical effects of the present invention, the system of the present invention and a comparative system that only uses feedforward control of a single ammonia injection channel are respectively operated under the same working conditions. When it is predicted that the inlet flue gas CO concentration drops to 2600 mg / Nm 3 and the flue gas temperature will drop below the 260 °C threshold in 6 minutes, the heat supplement switching sub-logic unit 4.2 of the present invention calculates the opening of the gas regulating valve of the hot blast stove to be 15% through the gas consumption prediction model, and starts the heat supplement in advance. Since the heat supplement process is seamlessly connected with the natural cooling process of the flue gas, the fluctuation of the inlet flue gas temperature of the medium-temperature SCR reactor is controlled within ±3 °C and always maintained in the optimal reaction range of 280 - 290 °C; while the comparative system lacks the flue gas temperature prediction feedforward and only relies on feedback heat supplement, the flue gas temperature fluctuates up to ±12 °C, and the instantaneous minimum temperature drops to 252 °C, resulting in an instantaneous decrease in the denitration efficiency.
[0050] In terms of ammonia injection control, the short-term learning channel of the long-term and short-term self-learning module of the present invention captures the inlet NO xThe instantaneous fluctuations in concentration, coupled with the long-term learning channel's 24-hour cycle to study the slow decline in catalyst activity, and the weighted fusion of the dual-channel signals, output a precise ammonia injection feedforward value. This improves ammonia injection control accuracy by over 70% compared to the control system using only short-term feedforward. Through the synergistic effect of the feedback and correction loops, ammonia slip is stably controlled below 2 ppm, while the control system experiences instantaneous peak ammonia slip of up to 8 ppm during sudden load changes.
[0051] After three months of continuous operation, the NO at the outlet of the mesophilic SCR reactor... x The average concentration was 24.6 mg / Nm³. 3 The number of instantaneous exceedances has been reduced to 0; the average gas consumption of the hot blast stove has been reduced by 90%, and the comprehensive energy consumption of the denitrification system has been reduced by 18%; the catalyst operating pressure difference is stable, and its service life is expected to be extended by more than 20%.
[0052] The above comparison results show that the dual-channel parallel feedforward architecture and long-short-term self-learning mechanism proposed in this invention have produced a significantly better synergistic technical effect than the existing single-channel feedforward technology in the control of medium-temperature SCR denitrification, and have fully achieved the purpose of the invention.
Claims
1. A smart collaborative control system for mesothermal selective catalytic reduction denitrification, characterized in that, It includes a multi-dimensional full-process perception module (1), a full-process working condition prediction module (2), a multi-objective collaborative optimization module (3), a collaborative control execution module, and an execution layer module (6) that are connected in sequence. The collaborative control execution module includes a flue gas temperature control module (4) and a precision ammonia injection collaborative control module (5) connected in parallel communication. The feedback output terminal of the execution layer module (6) is connected to the feedback input terminal of the multi-dimensional full-process perception module (1) to form a closed-loop control link for the entire process of medium-temperature SCR denitrification. The full-process operating condition prediction module (2) includes a historical operating database unit (2.1), a Long Short-Term Memory (LSTM) time series prediction unit (2.2), and a feedforward control signal output unit (2.3). The historical operating database unit (2.1) has a built-in offline learning module, which is used to train, generate, and update the prediction model based on historical data of the full operating conditions of the medium-temperature SCR denitrification system. The LSTM time series prediction unit (2.2) is an online learning module, which is used to update the model parameters in real time and predict the inlet flue gas temperature and NO of the medium-temperature SCR reactor in advance. x The changing trends of key operating parameters such as concentration, CO concentration, and flue gas flow rate, along with the prediction time, enable both flue gas temperature regulation and ammonia injection regulation to have advanced response capabilities. The output of the full-process operating condition prediction module (2) is connected to the input of the flue gas temperature control module (4) and the precise ammonia injection collaborative control module (5) respectively. It is used to synchronously output feedforward prediction signals to the flue gas temperature regulation and ammonia injection regulation before medium temperature SCR denitrification, so that the flue gas temperature regulation and ammonia injection regulation cooperate with each other in time to form a coordinated feedforward control of flue gas temperature and ammonia injection. The output of the multi-objective collaborative optimization module (3) is connected to the input of the flue gas temperature control module (4) and the precision ammonia injection collaborative control module (5) respectively, and is used to output the optimized operating parameter setting values. The flue gas temperature control module (4) is located at the inlet of the medium-temperature SCR reactor and includes a self-heating control sub-logic unit (4.1) and a supplementary heat switching sub-logic unit (4.2). The supplementary heat switching sub-logic unit (4.2) has a built-in gas consumption prediction model and is used to receive the flue gas temperature prediction signal output by the full-process operating condition prediction module (2). When it is predicted that the flue gas temperature at the inlet of the medium-temperature SCR reactor will be lower than the lower limit temperature of the catalyst activity window, the required gas consumption is calculated in real time, and the opening prediction value of the hot air furnace gas regulating valve is output in advance to achieve non-disruptive supplementary heat switching. The precise ammonia injection collaborative control module (5) is arranged in the inlet flue of the medium-temperature SCR reactor, including a feedforward loop unit (5.1), a feedback loop unit (5.2) and a correction loop unit (5.3); the feedforward loop unit (5.1) has a built-in long and short-term self-learning module, which is used to simultaneously collect minute-level inlet flue gas parameter mutation data and daily medium-temperature SCR operation trend data, and calculate the ammonia injection feedforward value.
2. The intelligent collaborative control system according to claim 1, characterized in that, The multi-objective collaborative optimization module (3) includes a rigid constraint boundary element (3.1), a multi-objective global optimization element (3.2), and an optimal operating parameter solution element (3.3); the multi-objective global optimization element (3.2) adopts a particle swarm optimization algorithm to determine the NO at the outlet of the medium-temperature SCR denitrification system. x The concentration and ammonia escape concentration meet the ultra-low emission limit as rigid constraints, and the optimization objective is to achieve the lowest comprehensive energy consumption of the denitrification system. The optimal operating parameter solution unit (3.3) solves the optimal parameter combination of the inlet flue gas temperature setpoint of the medium temperature SCR reactor, the opening degree of the ammonia injection regulating valve, and the soot blowing sequence of the flue gas-flue gas heat exchanger based on the output results of the particle swarm algorithm.
3. The intelligent collaborative control system according to claim 1, characterized in that, The lower limit temperature of the catalyst activity window of the supplementary heating switching sub-logic unit (4.2) is 260℃; the prediction duration of the LSTM time series prediction unit (2.2) is 5-10 minutes, so that the flue gas temperature regulation and ammonia injection regulation have a 5-10 minute advance response capability.
4. The intelligent collaborative control system according to claim 1, characterized in that, The long and short-term self-learning modules in the feedforward loop unit (5.1) include a short-term learning channel and a long-term learning channel. The short-term learning channel is used to capture the entry NO in the first cycle. x The instantaneous fluctuations in concentration and flue gas flow rate are tracked by the long-term learning channel, which is used to track the changing trends of catalyst activity decay and equipment characteristic drift in the second cycle. The first cycle is on the minute level, and the second cycle is on the day level. The outputs of the short-term learning channel and the long-term learning channel are weighted and fused to generate the ammonia injection feedforward value.
5. The intelligent collaborative control system according to claim 1, characterized in that, The offline learning module built into the historical operation database unit (2.1) retrains and updates the basic prediction model based on the newly added historical data of the entire operation of medium-temperature SCR denitrification in the first update cycle; the LSTM time series prediction unit (2.2) uses real-time operation data to fine-tune the model online in the second update cycle; the duration of the first update cycle is longer than the duration of the second update cycle.
6. The intelligent collaborative control system according to claim 1, characterized in that, The multi-dimensional full-process perception module (1) is distributed along the medium-temperature SCR denitrification process route, including a distributed operating condition measurement point subunit (1.1), a multi-point flue gas automatic monitoring system (CEMS) component (1.2), an equipment status sensor subunit (1.3), and a measurement point fault diagnosis and redundancy switching subunit (1.4); the execution layer module (6) includes a flue gas temperature control execution unit (6.1) and an ammonia injection system execution unit (6.2).
7. The intelligent collaborative control system according to claim 1, characterized in that, The feedforward prediction signal output by the full-process operating condition prediction module (2) simultaneously drives the flue gas temperature control module (4) to perform flue gas temperature feedforward adjustment and the precise ammonia injection coordinated control module (5) to perform ammonia injection feedforward adjustment. The responses of both control channels are ahead of the measured changes in the inlet parameters of the mesophilic SCR reactor. x When the load fluctuates, the flue gas temperature is adjusted in advance to create efficient active temperature conditions for the SCR catalyst, and the ammonia injection is adjusted synchronously to match the ammonia-nitrogen molar ratio, forming a time-sequential synergistic control.
8. The intelligent collaborative control system according to claim 1, characterized in that, The system is applied to the medium-temperature SCR denitrification process of sintering flue gas in steel. After semi-dry desulfurization, the flue gas enters the medium-temperature SCR reactor after being reheated by the flue gas-flue gas heat exchanger and the hot blast stove. The flue gas temperature control module (4) is used to adjust the opening of the gas regulating valve of the hot blast stove.
9. A method for intelligent synergistic control of mesophilic selective catalytic reduction denitrification, implemented based on the system described in any one of claims 1-8, characterized in that, Includes the following steps: S1: Collect the full-process operation data of medium-temperature SCR denitrification through the multi-dimensional full-process perception module (1) to complete data preprocessing and measurement point redundancy switching; S2: The full-process operating condition prediction module (2) predicts the parameter change trend of key operating conditions at the inlet of the medium-temperature SCR reactor through an offline + online dual learning mechanism, and outputs feedforward control signals to the two channels of the flue gas temperature control module (4) and the precision ammonia injection collaborative control module (5) to realize the collaborative feedforward control of flue gas temperature and ammonia injection. S3: The multi-objective collaborative optimization module (3) uses the particle swarm algorithm to perform global optimization, solves the optimal combination of operating parameters for the medium-temperature SCR denitrification reaction, and sends the optimized flue gas temperature setpoint and ammonia injection opening setpoint to the flue gas temperature control module (4) and the precision ammonia injection collaborative control module (5), respectively. S4: Flue gas temperature control module (4) adjusts the flue gas temperature to the optimal activity temperature window of the medium-temperature SCR catalyst based on the feedforward control signal and the optimized flue gas temperature setpoint. Meanwhile, the precision ammonia injection collaborative control module (5) performs long and short-term self-learning feedforward + feedback + correction collaborative ammonia injection control based on the feedforward control signal and the optimized ammonia injection opening setting value. S5: The execution layer module (6) executes control commands, adjusts the operation of the denitrification system and the induced draft fan, and sends the operation feedback data back to the multi-dimensional full-process perception module (1) to form a full-process closed-loop control.
10. The intelligent cooperative control method according to claim 9, characterized in that, The offline + online dual learning mechanism in step S2 is as follows: In the first update cycle, the LSTM base model is trained offline using the newly added historical running data and the model weights are updated. The model is fine-tuned online using real-time running data in the second update cycle. The first update cycle is longer than the second update cycle, so that the model can continuously adapt to catalyst activity decay and operating condition drift. In step S4, the supplementary heating switching sub-logic unit (4.2) of the flue gas temperature control module (4) receives the flue gas temperature prediction signal. When it is predicted that the flue gas temperature at the reactor inlet will fall below the lower limit temperature of the catalyst activity window within a preset time, the predicted value of the opening of the hot air furnace gas regulating valve is output in advance, so that the supplementary heating process and the flue gas cooling process are seamlessly connected.