Waste incineration flue gas multi-pollutant cooperative control system based on artificial intelligence

Abstract

The invention relates to the technical field of flue gas pollutant control, and discloses a waste incineration flue gas multi-pollutant cooperative control system based on artificial intelligence, which comprises a data acquisition module, a preprocessing module, an LSTM prediction module, a cooperative control module and an execution module. The acquisition module is used for acquiring DCS / PLC data in a millisecond level according to an OPC protocol; the preprocessing module divides a data set according to a ratio of 7: 2: 1 after performing 3 sigma exception removal, forward filling value compensation and Z-Score standardization; the LSTM module uses MSE as a loss function training model and predicts the pollutant concentration through Pearson's screening features; the cooperative control module comprises a deacidification cascade PID (with amplitude limiting / override) and denitration dual-objective function + PID, and PID parameters can be set on line; the system adopts a containerized architecture, so that the pollutant concentration can be pre-judged in advance, and the manual dependence is reduced.

Description

Technical Field

[0001] This application relates to the field of flue gas pollutant control technology, and in particular to an artificial intelligence-based multi-pollutant collaborative control system for waste incineration flue gas. Background Technology

[0002] The incineration of municipal solid waste releases a variety of pollutants, mainly including sulfur dioxide, hydrogen chloride, nitrogen oxides, particulate matter, and dioxins. Currently, there are still shortcomings in pollutant monitoring and intelligent control.

[0003] Currently, incineration plants mainly rely on Continuous Emission Monitoring Systems (CEMS) to monitor pollutants in real time and manually adjust the operation of purification equipment based on the monitoring results. However, this model has several limitations: First, to achieve ultra-low emissions of flue gas pollutants, effective coordinated control is required across multiple stages, including combustion, detection, and purification absorption. Under traditional control technologies, operators mainly rely on personal experience to adjust the incinerator. This method is highly dependent on the skill level of the operators, and the varying skill levels of different personnel can lead to large fluctuations in operating conditions, thus affecting the operational stability of the incinerator and pollutant emissions. Second, CEMS monitoring has an inherent lag. Fluctuations in combustion conditions caused by the complex composition of waste can further amplify the lag effect, often resulting in excessive pollutant emissions or over-addition of materials such as deacidifying agents and denitrifying agents, which increases operating costs and may cause secondary problems such as equipment blockage. Summary of the Invention

[0004] The purpose of this invention is to provide an artificial intelligence-based multi-pollutant collaborative control system for waste incineration flue gas, which can achieve synergistic control of HCl, SO2, and NOx, ensure stable and compliant pollutant emissions, and reduce the consumption of deacidifying agents (lime slurry) and denitrifying agents (urea), thereby reducing human intervention.

[0005] The technical solution of the multi-pollutant collaborative control system for waste incineration flue gas based on artificial intelligence provided in this application is as follows:

[0006] An AI-based multi-pollutant collaborative control system for waste incineration flue gas includes a data acquisition module for acquiring operational data from the DCS / PLC system of the waste incinerator at millisecond-level frequency via the OPC protocol. The operational data includes furnace temperature, excess air coefficient, primary / secondary air ratio, flue gas oxygen content, urea solution flow rate, lime slurry flow rate, boiler main steam parameters, and HCl, SO2, and NOx concentration data monitored by CEMS.

[0007] The data preprocessing module, which is connected in communication with the data acquisition module, is used to sort the acquired running data by time series, identify and remove outliers using the 3σ criterion, fill in missing values ​​using the forward imputation method, standardize the data using the Z-Score standardization method, and divide the standardized data into training set, validation set and test set in a ratio of 7:2:1.

[0008] The LSTM pollutant prediction model module is communicatively connected to the data preprocessing module and includes a model training server and a model inference server. The model training server constructs an LSTM time series prediction model based on the training set, and the LSTM time series prediction model uses MSE as the loss function.

[0009] The model inference server receives real-time operating data from the DCS and calls the trained LSTM time series prediction model to output the predicted concentration values ​​of HCl, SO2, and NOx.

[0010] The multi-pollutant collaborative control module is communicatively connected to the LSTM pollutant prediction model module, and includes a deacidification control submodule and a denitrification control submodule;

[0011] The execution module, which is communicatively connected to the multi-pollutant collaborative control module, includes a lime slurry flow regulating valve, a urea solution flow regulating valve, an SCR spray gun regulating valve, and an atomizer, and is used to adjust the equipment operating parameters according to the output instructions of the multi-pollutant collaborative control module.

[0012] Preferably, in the data preprocessing module, the Z-Score standardized mean and standard deviation are stored in a dedicated data storage file, and the validation set and test set are standardized using the same mean and standard deviation as the training set; in the 3σ criterion, outliers are defined as values ​​that deviate from the data mean by 3 times the standard deviation.

[0013] Preferably, before training, the LSTM pollutant prediction model module uses the Pearson correlation coefficient method to screen input features, selecting operating parameters that have a significant linear correlation with the concentrations of HCl, SO2, and NOx as model input features. The significance of the correlation is determined based on the historical operating data of the specific project. The operating parameters that are determined to be significantly correlated after analysis indicate that the higher the correlation between the operating parameter and the concentrations of HCl, SO2, and NOx, the higher the correlation. Typical highly correlated operating parameters include, but are not limited to: incinerator outlet flue gas temperature, urea solution flow rate, primary air fan current, and evaporator inlet flue gas temperature.

[0014] Preferably, the deacidification control submodule adopts a cascaded PID collaborative control strategy, including a main PID control layer and a secondary PID control layer. The main PID control layer is equipped with two PID controllers, which are setpoints for the target emission concentrations of HCl and SO2, respectively, and the predicted concentrations of HCl and SO2 output by the LSTM pollutant prediction model module as process variables, outputting the lime slurry demand value. The two lime slurry demand values ​​of the main PID control layer are multiplied by weights K1 and K2 respectively and then summed to obtain the total lime slurry calculation amount, where K1 is the HCl removal adaptation weight coefficient and K2 is the SO2 removal adaptation weight coefficient. The secondary PID control layer uses the total lime slurry calculation amount as the setpoint and the actual lime slurry flow rate as the process variable, outputting commands to control the lime slurry flow regulating valve.

[0015] Preferably, the deacidification control submodule further includes a limiting module and an overshoot module. The limiting module is used to constrain the upper and lower limits of the total lime slurry accumulation value to avoid system disorder caused by the output of the control module's calculation results being too slow or too fast under special working conditions. The overshoot module is used to intervene in advance to amplify the total lime slurry accumulation value to avoid system sluggish response caused by the output range of the control module's calculation results being too narrow under special working conditions.

[0016] Preferably, the denitrification control submodule constructs a dual objective function, including a cost function and an environmental indicator function; the cost function is:

[0017] During the denitrification process, NO accounts for 95% of the total NOx volume concentration, and NO2 accounts for 5% of the total NOx volume concentration. To simplify the calculation, only NO is considered. In the formula... To measure NOx concentration, Let E be the predicted NOx emission concentration, Qy be the flue gas flow rate, and 0.8 be the ammonia nitrogen equivalent molar ratio; the environmental protection index function is: E NOX =f(Q), where Q is the ammonia flow rate and f is the NOx prediction model. The final objective function of the denitrification control submodule is:

[0018]

[0019] α is the weighting coefficient of the cost function, and β is the weighting coefficient of the environmental protection function.

[0020] Preferably, the denitrification control submodule adopts PID control logic. The main PID controller uses the NOx target emission concentration as the set value and the NOx predicted concentration output by the LSTM pollutant prediction model module as the process variable to output the ammonia water demand flow rate. The auxiliary PID controller uses the ammonia water demand flow rate as the set value and the actual ammonia water flow rate as the process variable to output commands to control the urea solution flow regulating valve or the SCR spray gun regulating valve.

[0021] Preferably, the model training server and the model inference server adopt a distributed containerized architecture, and the functional modules are encapsulated as microservices through Docker containers. The containers interact with each other based on the HTTP protocol and are compatible with OPC, RTSP industrial protocols and DCS system interfaces.

[0022] Preferably, the multi-pollutant collaborative control module further includes a control logic optimization unit, which is used to dynamically correct the proportional coefficient and integral time of the PID controller based on the deviation between the pollutant concentration monitored in real time by CEMS and the concentration predicted by LSTM, so as to realize online tuning of PID parameters.

[0023] In summary, this application includes the following beneficial technical effects:

[0024] 1. The system can predict pollutant concentrations in advance, reducing reliance on manual intervention. It collects incinerator operation data and pollutant concentration data in real time, and also removes abnormal data and completes missing data to ensure data reliability. Furthermore, the model calculates the concentrations of HCl, SO2, and NOx 10 seconds in advance, solving the problem of traditional monitoring lag. Compared to relying on manual experience in the past, the system now relies entirely on real data for control, enabling timely responses to changes in pollutant concentrations and avoiding emissions exceeding standards due to slow reaction. This allows for more accurate control of multiple pollutants in accordance with actual operating conditions.

[0025] 2. It can precisely control pollutant emissions while saving on reagent costs. During deacidification, it prioritizes ensuring compliance with stricter HCl emission standards, precisely calculating the amount of lime slurry to be used. During denitrification, it considers both environmental compliance and ammonia costs, avoiding waste from excessive dosage or exceeding limits due to insufficient dosage. It can also precisely control reagent flow, reducing the risk of equipment blockage, thus complying with environmental regulations while reducing reagent costs, balancing pollutant control effectiveness with operational economy.

[0026] 3. Energy-saving and adaptable to different incineration processes. The system utilizes purified high-temperature flue gas to preheat new flue gas, achieving high heat exchange efficiency and significantly saving electricity; electric heating is activated only when the temperature is insufficient, maintaining the catalyst's efficient operation. Furthermore, the system is divided into independent functional modules, supporting multiple industrial communication methods, eliminating the need for separate modifications for different waste incineration plants and adapting to various processes, thus enhancing its applicability in diverse scenarios. Attached Figure Description

[0027] Figure 1 This is a diagram of the overall system architecture of the present invention.

[0028] Figure 2 This is a flowchart of the training process for the LSTM time series prediction model of this invention.

[0029] Figure 3 This is the control logic diagram of the deacidification synergistic control module of the present invention.

[0030] Figure 4 This is the control logic diagram of the denitrification synergistic control module of the present invention.

[0031] Figure 5 This is a schematic diagram of the PID control structure of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0033] Reference Figure 1-5 This invention provides an artificial intelligence-based multi-pollutant collaborative control system for waste incineration flue gas, including a data acquisition module for acquiring operational data from the DCS / PLC system of the waste incinerator at a millisecond frequency via the OPC protocol. The operational data includes furnace temperature, excess air coefficient, primary / secondary air ratio, flue gas oxygen content, urea solution flow rate, lime slurry flow rate, boiler main steam parameters, and HCl, SO2, and NOx concentration data monitored by CEMS.

[0034] The data preprocessing module communicates with the data acquisition module and is used to sort the acquired running data by time series, identify and remove outliers using the 3σ criterion, fill in missing values ​​using the forward imputation method, standardize the data using the Z-Score standardization method, and divide the standardized data into training set, validation set and test set in a ratio of 7:2:1.

[0035] The LSTM pollutant prediction model module communicates with the data preprocessing module and includes a model training server and a model inference server. The model training server builds an LSTM time series prediction model based on the training set, and the LSTM time series prediction model uses MSE as the loss function.

[0036] The model inference server receives real-time operating data from the DCS and calls the trained LSTM time series prediction model to output the predicted concentration values ​​of HCl, SO2, and NOx.

[0037] The multi-pollutant collaborative control module communicates with the LSTM pollutant prediction model module, including a deacidification control submodule and a denitrification control submodule;

[0038] The execution module, which communicates with the multi-pollutant collaborative control module, includes a lime slurry flow regulating valve, a urea solution flow regulating valve, an SCR spray gun regulating valve, and an atomizer. It is used to adjust the equipment operating parameters according to the output instructions of the multi-pollutant collaborative control module.

[0039] Using the OPC protocol, key combustion parameters (such as furnace temperature and excess air coefficient), reagent addition data (such as urea solution flow rate), and HCl, SO2, and NOx concentrations monitored by CEMS are collected in real time from the DCS / PLC system to ensure complete data acquisition throughout the entire process. The data is then organized chronologically, outliers are eliminated using the 3σ method, missing values ​​are filled, and Z-score normalization is applied. Finally, the data is divided into training, validation, and test sets in a 7:2:1 ratio to ensure data quality. For the LSTM model, the training server uses mean squared error (MSE) as the loss function and Adam as the optimizer (learning rate set to 0.001). The system is designed with a historical data length of 60 and a prediction step size of 10. After 200 rounds of training, the parameters are adjusted using a validation set to prevent overfitting. The inference server receives real-time data and outputs the predicted pollutant concentration. The control module is divided into two parts: acid removal and denitrification, which are adjusted accordingly. The execution module converts the control commands into actual adjustments to the equipment parameters through regulating valves. The main advantage of this design is that the entire process is based on real operating data, avoiding the subjectivity of traditional control methods that rely on human experience. At the same time, the LSTM model can predict pollutant concentrations in advance, alleviating the problem of CEMS monitoring delay. The modular structure also makes the system more adaptable to different incineration processes, improving its overall applicability.

[0040] In a preferred embodiment, in the data preprocessing module, the Z-Score standardized mean and standard deviation are stored in a dedicated data storage file, and the validation set and test set are standardized using the same mean and standard deviation as the training set; in the 3σ criterion, outliers are defined as values ​​that deviate from the data mean by 3 times the standard deviation.

[0041] In the data preprocessing section, Z-Score standardization requires calculating the mean and standard deviation of the training data and storing these parameters in a dedicated data storage file. Validation and test data are then processed directly using the parameters stored in this file to ensure all data uses the same scaling standard. Simultaneously, the 3σ method is used to identify outliers, classifying data that deviate from the mean by more than three times the standard deviation as outliers. Missing values ​​and outliers are imputed using the previous valid value.

[0042] This approach ensures more objective model evaluation, reduces the impact of extreme data on model training, effectively improves data quality, and provides a more reliable data foundation for subsequent model training.

[0043] In a preferred embodiment, before training, the LSTM pollutant prediction model module uses the Pearson correlation coefficient method to screen input features, selecting operating parameters that have a significant linear correlation with the concentrations of HCl, SO2, and NOx as model input features. The significance of the correlation is determined based on the historical operating data of the specific project. Operating parameters that are determined to be significantly correlated after analysis indicate that the higher the correlation between the operating parameter and the concentrations of HCl, SO2, and NOx, the higher the correlation. Typical highly correlated operating parameters include, but are not limited to: incinerator outlet flue gas temperature, urea solution flow rate, primary air fan current, and evaporator inlet flue gas temperature.

[0044] The linear relationship between various operating parameters and the concentrations of HCl, SO2, and NOx was analyzed using the Pearson correlation coefficient method. Parameters with correlation coefficients closer to 1 were selected as input features, such as incinerator outlet flue gas temperature, urea solution flow rate, primary air fan current, and evaporator inlet flue gas temperature. Parameters with weak correlations were discarded to simplify the model structure. This method effectively reduces the number of input variables, shortens computation time, and reduces memory usage. It also highlights key factors affecting pollutant concentrations and avoids interference from irrelevant variables, thereby improving the accuracy of model predictions and making the predicted results for HCl and SO2 pollutants more consistent with actual control needs.

[0045] In a preferred embodiment, the deacidification control submodule employs a cascaded PID collaborative control strategy, including a main PID control layer and a secondary PID control layer. The main PID control layer sets two PID controllers, with the target emission concentrations of HCl and SO2 as setpoints, and the predicted concentrations of HCl and SO2 output by the LSTM pollutant prediction model module as process variables, outputting the lime slurry demand value. The two lime slurry demand values ​​of the main PID control layer are multiplied by weights K1 and K2 respectively and then summed to obtain the total lime slurry calculation amount, where K1 is the HCl removal adaptation weight coefficient and K2 is the SO2 removal adaptation weight coefficient. The secondary PID control layer uses the total lime slurry calculation amount as the setpoint and the actual lime slurry flow rate as the process variable, outputting commands to control the lime slurry flow regulating valve.

[0046] The deacidification control section employs a two-stage adjustment strategy. The first stage uses two controllers, one taking the target emission concentration and the other the model-predicted concentration as input, to calculate the required amount of lime slurry. The second stage uses this total required amount as the target and the actual lime slurry flow rate as feedback to adjust the valve opening. This method can reconcile the inconsistency in HCl and SO2 concentration changes, prioritizing compliance with the stricter HCl emission requirements, while using predicted values ​​instead of lagging measurements to improve the efficiency of the deacidification agent.

[0047] In a preferred embodiment, the deacidification control submodule further includes a limiting module and an overshoot module. The limiting module is used to constrain the upper and lower limits of the total lime slurry accumulation value to prevent the system from becoming disordered due to the output of the control module's calculation results being too slow or too fast under special working conditions. The overshoot module is used to intervene in advance to amplify the total lime slurry accumulation value to prevent the system from becoming sluggish due to the output range of the control module's calculation results being too narrow under special working conditions.

[0048] The inclusion of limiting and override modules effectively improves the stability and response efficiency of the deacidification control submodule under complex operating conditions. The limiting module, by constraining the upper and lower limits of the total lime slurry accumulation value, prevents both reagent waste and equipment blockage caused by excessive lime slurry addition, and avoids excessive HCl and SO2 emissions due to insufficient dosage. The override module, addressing the issue of a narrow control output range, amplifies the accumulation value through early intervention, quickly matching sudden increases in pollutant concentrations and resolving the system's sluggish response caused by hysteresis, ensuring timely and effective control.

[0049] In a preferred embodiment, the denitrification control submodule constructs a dual-objective function, including a cost function and an environmental performance indicator function; the cost function is:

[0050] During the denitrification process, NO accounts for 95% of the total NOx volume concentration, and NO2 accounts for 5% of the total NOx volume concentration. To simplify the calculation, only NO is considered. In the formula... To measure NOx concentration, Let E be the predicted NOx emission concentration, Qy be the flue gas flow rate, and 0.8 be the ammonia nitrogen equivalent molar ratio; the environmental indicator function is: E NOX =f(Q), where Q is the ammonia flow rate and f is the NOx prediction model. The final objective function of the denitrification control submodule is:

[0051] α is the weighting coefficient of the cost function, and β is the weighting coefficient of the environmental protection function.

[0052] The denitrification control system considers both operating costs and environmental performance indicators. The cost function calculates ammonia usage based on the set concentration, predicted concentration, and flue gas flow rate, while the environmental function assesses whether emissions meet standards using a predictive model. Finally, the two are weighted and combined into a single overall objective function. This method overcomes the problems of excessive usage leading to waste or insufficient usage resulting in excessive emissions, ensuring compliance while minimizing costs.

[0053] In a preferred embodiment, the denitrification control submodule adopts PID control logic. The main PID controller uses the target NOx emission concentration as the set value and the predicted NOx concentration output by the LSTM pollutant prediction model module as the process variable to output the ammonia water demand flow rate. The secondary PID controller uses the ammonia water demand flow rate as the set value and the actual ammonia water flow rate as the process variable to output commands to control the urea solution flow regulating valve or the SCR spray gun regulating valve.

[0054] The main controller of the denitrification control section calculates the required ammonia flow rate using the target concentration as the setpoint and the predicted concentration as feedback. The secondary controller adjusts the valve opening based on this required flow rate and the actual flow rate. This strategy mitigates system response delay through predictive control, reduces concentration fluctuations when operating conditions change, and the secondary controller precisely controls the flow rate, preventing both insufficient flow leading to poor denitrification and excessive flow causing equipment blockage.

[0055] In a preferred embodiment, the model training server and the model inference server adopt a distributed containerized architecture, and the functional modules are encapsulated as microservices through Docker containers. The containers interact with each other based on the HTTP protocol and are compatible with OPC, RTSP industrial protocols and DCS system interfaces.

[0056] The training and prediction servers adopt a distributed container design, with each function encapsulated as an independent module. They exchange data through a common protocol, support multiple industrial communication methods, and can be deployed separately to meet different performance requirements.

[0057] In a preferred embodiment, the multi-pollutant collaborative control module further includes a control logic optimization unit, which dynamically corrects the proportional coefficient and integral time of the PID controller based on the deviation between the pollutant concentration monitored in real time by CEMS and the concentration predicted by LSTM, thereby achieving online tuning of the PID parameters.

[0058] The control module compares the deviation between the measured concentration and the predicted concentration in real time. When the deviation is small, it maintains the original parameters, and when the deviation is large, it automatically adjusts the proportional and integral parameters. For example, when the measured concentration is higher than the predicted concentration, the proportional coefficient is increased and the integral time is decreased. The new parameters are sent to the controller in real time to achieve automatic parameter optimization.

[0059] The foregoing description, with reference to preferred embodiments, illustrates an exemplary implementation of an artificial intelligence-based multi-pollutant collaborative control system for waste incineration flue gas provided by this disclosure. However, those skilled in the art will understand that various modifications and alterations can be made to the above specific embodiments without departing from the spirit of this disclosure, and various combinations can be made to the various technical features and structures proposed in this disclosure without exceeding the protection scope of this disclosure, the protection scope of which is determined by the appended claims.

Claims

1. A multi-pollutant collaborative control system for waste incineration flue gas based on artificial intelligence, characterized in that: It includes a data acquisition module, which is used to collect operating data from the DCS / PLC system of the waste incinerator at a millisecond frequency via the OPC protocol. The operating data includes furnace temperature, excess air coefficient, primary / secondary air ratio, flue gas oxygen content, urea solution flow rate, lime slurry flow rate, boiler main steam parameters, and HCl, SO2, and NOx concentration data monitored by CEMS. The data preprocessing module, which is connected in communication with the data acquisition module, is used to sort the acquired running data by time series, identify and remove outliers using the 3σ criterion, fill in missing values ​​using the forward imputation method, standardize the data using the Z-Score standardization method, and divide the standardized data into training set, validation set and test set in a ratio of 7:2:

1. The LSTM pollutant prediction model module is communicatively connected to the data preprocessing module and includes a model training server and a model inference server. The model training server constructs an LSTM time series prediction model based on the training set, and the LSTM time series prediction model uses MSE as the loss function. The model inference server receives real-time operating data from the DCS and calls the trained LSTM time series prediction model to output the predicted concentration values ​​of HCl, SO2, and NOx. The multi-pollutant collaborative control module is communicatively connected to the LSTM pollutant prediction model module, and includes a deacidification control submodule and a denitrification control submodule; The execution module, which is communicatively connected to the multi-pollutant collaborative control module, includes a lime slurry flow regulating valve, a urea solution flow regulating valve, an SCR spray gun regulating valve, and an atomizer, and is used to adjust the equipment operating parameters according to the output instructions of the multi-pollutant collaborative control module.

2. The multi-pollutant collaborative control system for waste incineration flue gas based on artificial intelligence according to claim 1, characterized in that: In the data preprocessing module, the Z-Score standardized mean and standard deviation are stored in a dedicated data storage file, and the validation set and test set are standardized using the same mean and standard deviation as the training set; in the 3σ criterion, outliers are defined as values ​​that deviate from the data mean by 3 times the standard deviation.

3. The multi-pollutant collaborative control system for waste incineration flue gas based on artificial intelligence according to claim 1, characterized in that: Before training, the LSTM pollutant prediction model module uses the Pearson correlation coefficient method to screen input features and selects operating parameters that have a significant linear correlation with the concentrations of HCl, SO2, and NOx as model input features. The significance of the correlation is determined based on the historical operating data of the specific project. The operating parameters that are determined to be significantly correlated after analysis indicate that the higher the correlation between the operating parameter and the concentrations of HCl, SO2, and NOx, the higher the correlation. Typical highly correlated operating parameters include, but are not limited to: incinerator outlet flue gas temperature, urea solution flow rate, primary air fan current, and evaporator inlet flue gas temperature.

4. The multi-pollutant collaborative control system for waste incineration flue gas based on artificial intelligence according to claim 1, characterized in that: The deacidification control submodule adopts a cascaded PID collaborative control strategy, including a main PID control layer and a secondary PID control layer. The main PID control layer sets two PID controllers, with the target emission concentrations of HCl and SO2 as setpoints, and the predicted concentrations of HCl and SO2 output by the LSTM pollutant prediction model module as process variables, outputting the lime slurry demand value. The two lime slurry demand values ​​of the main PID control layer are multiplied by weights K1 and K2 respectively and then summed to obtain the total lime slurry calculation amount, where K1 is the HCl removal adaptation weight coefficient and K2 is the SO2 removal adaptation weight coefficient. The secondary PID control layer uses the total lime slurry calculation amount as the setpoint and the actual lime slurry flow rate as the process variable, outputting commands to control the lime slurry flow regulating valve.

5. The multi-pollutant collaborative control system for waste incineration flue gas based on artificial intelligence according to claim 1, characterized in that: The deacidification control submodule also includes a limiting module and an override module. The limiting module is used to constrain the upper and lower limits of the total lime slurry accumulation value to avoid system disorder caused by the output of the control module's calculation results being too slow or too fast under special working conditions. The override module is used to intervene in advance to amplify the total lime slurry accumulation value to avoid system sluggish response caused by the output range of the control module's calculation results being too narrow under special working conditions.

6. The multi-pollutant collaborative control system for waste incineration flue gas based on artificial intelligence according to claim 1, characterized in that: The denitrification control submodule constructs a dual objective function, including a cost function and an environmental indicator function; the cost function is: During the denitrification process, NO accounts for 95% of the total NOx volume concentration, and NO2 accounts for 5% of the total NOx volume concentration. To simplify the calculation, only NO is considered. In the formula... To measure NOx concentration, Let E be the predicted NOx emission concentration, Qy be the flue gas flow rate, and 0.8 be the ammonia nitrogen equivalent molar ratio; the environmental protection index function is: E NOX =f(Q), where Q is the ammonia flow rate and f is the NOx prediction model. The final objective function of the denitrification control submodule is: α is the weighting coefficient of the cost function, and β is the weighting coefficient of the environmental protection function.

7. The multi-pollutant collaborative control system for waste incineration flue gas based on artificial intelligence according to claim 1, characterized in that: The denitrification control submodule adopts PID control logic. The main PID controller uses the target NOx emission concentration as the set value and the NOx predicted concentration output by the LSTM pollutant prediction model module as the process variable, and outputs the ammonia water demand flow rate. The auxiliary PID controller uses the ammonia water demand flow rate as the set value and the actual ammonia water flow rate as the process variable, and outputs commands to control the urea solution flow regulating valve or the SCR spray gun regulating valve.

8. The multi-pollutant collaborative control system for waste incineration flue gas based on artificial intelligence according to claim 1, characterized in that: The model training server and model inference server adopt a distributed containerized architecture, and the functional modules are encapsulated as microservices through Docker containers. The containers interact with each other based on the HTTP protocol and are compatible with OPC, RTSP industrial protocols and DCS system interfaces.

9. The multi-pollutant collaborative control system for waste incineration flue gas based on artificial intelligence according to claim 1, characterized in that: The multi-pollutant collaborative control module also includes a control logic optimization unit, which dynamically corrects the proportional coefficient and integral time of the PID controller based on the deviation between the pollutant concentration monitored in real time by CEMS and the concentration predicted by LSTM, thereby achieving online tuning of the PID parameters.

Images