Method for improving denitration agent adding precision and controlling system response speed

Abstract

The invention relates to a method for improving the adding precision of a denitration agent and controlling the response speed of a system, and belongs to the technical field of sintering desulfurization and denitration system control methods. According to the technical scheme, system operation state parameters are obtained through a sensor system, and a data training set is prepared; a denitration agent consumption prediction model is constructed, a data training set is used for learning, and model control parameters are obtained through online operation; constructing a cascade PID (Proportion Integration Differentiation) regulator which takes the NOx concentration of a chimney outlet as a set value and takes the frequency of an ammonia water motor as an output value in the L1 system; model control parameters serve as controller feedforward and are used for L1 closed-loop control, and the running frequency of the ammonia water motor is adjusted. The method has the advantages that the control precision of denitration agent adding and the response speed of an actuator can be remarkably improved, the method is suitable for various industrial application scenes, and the control effect can be continuously optimized through a machine learning model.

Description

Technical Field

[0001] This invention relates to a method for improving the accuracy of denitrification agent addition and the response speed of the control system, belonging to the technical field of control methods for sintering desulfurization and denitrification systems. Background Technology

[0002] In the field of industrial automation, high-precision equipment control is key to improving production efficiency and reducing consumption. In traditional sintering desulfurization and denitrification processes, ammonia is added in advance. After the chemical reaction in the reactor, the ammonia is piped to the end of the chimney to obtain feedback on the nitrogen oxide content, which is then used to regulate the dosage of the denitrifying agent. Meanwhile, the nitrogen oxide content at the chimney outlet is closely related to various factors such as temperature, pressure, and catalyst activity during production. Traditional denitrifying agent dosage control methods are usually based on fixed control rules, such as simple PID (proportional-integral-derivative) process control. These methods are difficult to adapt to complex and changing operating environments with multivariable time lags, resulting in drawbacks such as low control accuracy and response speed, and high production costs. Summary of the Invention

[0003] The purpose of this invention is to provide a method for improving the accuracy of denitrifying agent dosing and the response speed of the control system. By employing big data on-site extraction that meets the characteristic values, selection and synthesis of feedforward signal positions, neural network modeling, and basic automation control technology, the method can significantly improve the control accuracy of denitrifying agent dosing and the response speed of the actuator. It is applicable to various industrial application scenarios and can continuously optimize the control effect through machine learning models, effectively solving the aforementioned problems existing in the background technology.

[0004] The technical solution of this invention is: a method for improving the accuracy of denitrification agent dosing and the response speed of the control system, comprising the following steps:

[0005] S1. Data advance acquisition: Obtain system operating status parameters through sensor system and prepare data training set;

[0006] S2. Model building: Construct a denitrification agent consumption prediction model. The denitrification agent consumption prediction model is obtained by machine learning training based on neural network algorithm and its model control parameters are initialized.

[0007] S3. Model learning and parameter generation: After processing the system operation status parameters in step S1, the parameters are fed into the denitrification agent consumption prediction model in step S2 for learning.

[0008] S4. Controller construction: Construct a cascade PID controller in the L1 system with the NOx concentration at the chimney outlet as the set value and the frequency of the ammonia water dispensing motor as the output value.

[0009] S5, the recommended dosage of denitrifying agent, serves as a feedforward parameter for the cascade PID controller, acting as a reference calculation data between controller stages. It participates in the L1 closed-loop control to adjust the operating parameters of the ammonia water motor, thereby achieving high-precision control based on model prediction.

[0010] In step S1, the collected system operating status parameters include the flue gas volume at the dust collector outlet, the NO content at the dust collector outlet, the NO2 content at the dust collector outlet, the temperature at the direct-fired furnace outlet, the flue gas pressure at the direct-fired furnace outlet, the NO content at the chimney outlet, the NO2 content at the chimney outlet, and the NH3 content at the chimney outlet.

[0011] In step S1, the system operating status parameters are collected at locations before the ammonia reactor, providing preliminary calculations for PID setting parameters. The calculated values ​​form the de facto system feedforward.

[0012] In step S4, the steps for constructing the cascaded PID controller are as follows:

[0013] S4.1 Constructing the pre-stage regulator

[0014] The pre-regulator is a nitrogen oxide regulator, set to the expected nitrogen oxide concentration in mg / Nm³. 3 The feedback value is the actual feedback value of nitrogen oxides, with the same unit as the set value, and a proportional-integral controller is used.

[0015] S4.2 Constructing the Post-Stage Regulator

[0016] The downstream regulator is a denitrifying agent flow regulator. The setpoint is the weighted calculated value of the upstream regulator output and the model control parameters, in kg / h. The feedback value is the actual feedback value of the denitrifying agent, in the same unit as the setpoint. The regulator output is the denitrifying agent dosing device speed setpoint, in rpm. It uses a proportional-integral regulator.

[0017] The beneficial effects of this invention are: by employing big data on-site extraction that meets feature values, selection and synthesis of feedforward signal positions, neural network modeling, and basic automation control technology, it can significantly improve the control accuracy and actuator response speed of denitrifying agent addition, making it suitable for various industrial application scenarios, and the control effect can be continuously optimized through machine learning models. Attached Figure Description

[0018] Figure 1 This is a flowchart of the process of this invention;

[0019] Figure 2 This is a schematic diagram of the principle of the present invention;

[0020] Figure 3 This is a schematic diagram showing the location of the control model data acquisition and execution device of the present invention;

[0021] Figure 4 This is the signal transmission logic diagram of the control model of this invention;

[0022] Figure 5 This is a schematic diagram of the calculation process of the control model of the present invention;

[0023] In the diagram: 1. System operating status parameter acquisition location; 2. Feedback data source location; 3. Ammonia water motor; 4. Ammonia reactor; 5. Desulfurization tower; 6. Cyclone separator; 7. Bag filter; 8. Circulating ash silo; 9. Circulating ash conveyor; 10. Combustion furnace; 11. Cascade PID controller; 12. Chimney; 13. Flue gas inlet. Detailed Implementation

[0024] To make the purpose, technical solutions, and advantages of the invention's embodiments clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only a small part of the embodiments of the present invention, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0025] A method for improving the accuracy of denitrifying agent dosing and the response speed of the control system includes the following steps:

[0026] S1. Data advance acquisition: Obtain system operating status parameters through sensor system and prepare data training set;

[0027] S2. Model building: Construct a denitrification agent consumption prediction model. The denitrification agent consumption prediction model is obtained by machine learning training based on neural network algorithm and its model control parameters are initialized.

[0028] S3. Model learning and parameter generation: After processing the system operation status parameters in step S1, the parameters are fed into the denitrification agent consumption prediction model in step S2 for learning.

[0029] S4. Controller construction: Construct a cascade PID controller in the L1 system with the NOx concentration at the chimney outlet as the set value and the frequency of the ammonia water dispensing motor as the output value.

[0030] S5, the recommended dosage of denitrifying agent, serves as a feedforward parameter for the cascade PID controller, acting as a reference calculation data between controller stages. It participates in the L1 closed-loop control to adjust the operating parameters of the ammonia water motor, thereby achieving high-precision control based on model prediction.

[0031] In step S1, the collected system operating status parameters include the flue gas volume at the dust collector outlet, the NO content at the dust collector outlet, the NO2 content at the dust collector outlet, the temperature at the direct-fired furnace outlet, the flue gas pressure at the direct-fired furnace outlet, the NO content at the chimney outlet, the NO2 content at the chimney outlet, and the NH3 content at the chimney outlet.

[0032] In step S1, the system operating status parameters are collected at locations before the ammonia reactor, providing preliminary calculations for PID setting parameters. The calculated values ​​form the de facto system feedforward.

[0033] In step S4, the steps for constructing the cascaded PID controller are as follows:

[0034] S4.1 Constructing the pre-stage regulator

[0035] The pre-regulator is a nitrogen oxide regulator, set to the expected nitrogen oxide concentration in mg / Nm³. 3 The feedback value is the actual feedback value of nitrogen oxides, with the same unit as the set value, and a proportional-integral controller is used.

[0036] S4.2 Constructing the Post-Stage Regulator

[0037] The downstream regulator is a denitrifying agent flow regulator. The setpoint is the weighted calculated value of the upstream regulator output and the model control parameters, in kg / h. The feedback value is the actual feedback value of the denitrifying agent, in the same unit as the setpoint. The regulator output is the denitrifying agent dosing device speed setpoint, in rpm. It uses a proportional-integral regulator.

[0038] In practical applications, this invention aims to solve the problems of low control accuracy and insufficient response speed caused by the lag between the feedback data source and the actuator in current control technologies. It proposes a high-precision control method by combining big data AI modeling and basic automation technologies, including the following steps:

[0039] 1. Advanced Data Acquisition: The system acquires operating status parameters such as temperature, pressure, flow rate, flue gas NOx concentration, and NH3 content through a sensor system. The data acquisition location is at the front end of the ammonia reactor (SCR), enabling the prediction of consumption of the denitrification agent (ammonia water) before the denitrification operation.

[0040] 2. Model Construction: Construct a denitrification agent consumption prediction model. The prediction model is obtained by machine learning training based on a neural network algorithm, and its network parameters are initialized.

[0041] 3. Model learning and parameter generation: After processing the parameters described in step 1, feed them into the neural network model described in step 2 for learning;

[0042] 4. Regulator construction: Construct a cascade PID regulator in the L1 system with the NOx concentration at the chimney outlet as the set value and the frequency of the ammonia water dispensing motor as the output value;

[0043] 5. System Execution: The model established in step 2 is run online to calculate the recommended baseline value for the denitrification agent dosage. The regulator's operation is adjusted based on this recommended baseline value to achieve high-precision control. The recommended baseline value for the denitrification agent dosage serves as the reference parameter for the inter-stage coupling variables of the cascade PID regulator. Its calculation data is sampled physically before the denitrification agent reactor, allowing for advance calculation and parameter setting before regulator execution. This forms a feedforward parameter pre-setting mechanism, participating in the L1 closed-loop control to adjust the operating parameters of the ammonia water motor, achieving high-precision control based on model prediction.

[0044] Furthermore, the aforementioned data advance acquisition steps include:

[0045] 1.1 The data fully characterizes the consumption of denitrifying agent, and includes: (1) flue gas volume at the dust collector outlet, (2) NO content at the dust collector outlet, (3) NO2 content at the dust collector outlet, (4) temperature at the direct-fired furnace outlet, (5) flue gas pressure at the direct-fired furnace outlet, (6) NO content at the chimney outlet, (7) NO2 content at the chimney outlet, and (8) NH3 content at the chimney outlet.

[0046] 1.2 The data mentioned is the feature-refined data after manual screening and secondary machine processing following the initial coarse data collection on-site. Data collected in actual industrial processes often contains outliers, meaning that the collected data values ​​may fluctuate at certain times. Direct use of such data would undoubtedly cause significant interference to the system's identification process. Manual screening removes abnormal data that clearly does not conform to the production scenario, and the machine further performs zero-mean normalization on the raw data to achieve the effect of filtering and smoothing the data. The data collection employs segmented smoothing zero-mean normalization processing, and the specific algorithm is as follows:

[0047] (1)

[0048] (2)

[0049]

[0050] (3)

[0051] in:

[0052] Formula (1) is the first smoothing, Formula (2) is the second smoothing, and Formula (3) is the m-th smoothing.

[0053] m: number of smoothing cycles, p: sliding range;

[0054] Initial data, initial data length is n: data length.

[0055] Calculation in progress

[0056] Adjusting different values ​​of m and p will produce different smoothing effects.

[0057] Furthermore, step 2, model construction, includes:

[0058] The neural network is a feedforward-backpropagation neural network. It propagates the input signal forward and adjusts the network weights by propagating the error backward. The chain rule is used to calculate the gradient of the loss function with respect to the weights of each layer, thereby updating the network parameters to minimize the loss. The basic principle formula is as follows:

[0059] Inter-layer signal transmission:

[0060] Feedforward propagation:

[0061] Input layer to hidden layer:

[0062]

[0063] Where W(1) is the weight matrix, b(1) is the bias vector, and f(1) is the activation function.

[0064] Hidden layer to output layer:

[0065]

[0066] Where W(2) is the weight matrix, b(2) is the output layer bias vector, and f(2) is the output layer activation function.

[0067] Backpropagation:

[0068] Output error:

[0069]

[0070] Where y: the actual output. Derivative of the activation function of the output layer.

[0071] Hidden layer error:

[0072]

[0073] in, Transpose of the output layer weight matrix Derivative of hidden layer activation function Dot product.

[0074] Update weights and biases

[0075]

[0076]

[0077]

[0078]

[0079] in, Learning efficiency.

[0080] Furthermore, in step 3, model learning and parameter generation, the interface data format is as follows:

[0081] , .

[0082] in:

[0083] m: Number of groups of interface data.

[0084] Further, step 4 involves constructing a cascaded PID controller. The construction steps for a cascaded controller are as follows:

[0085] 4.1 Constructing the pre-stage regulator

[0086] The pre-regulator is a nitrogen oxide (NOx) regulator, set to the expected NOx concentration in mg / Nm³. 3 The feedback value is the actual feedback value of nitrogen oxides, with the same unit as the set value, and uses a proportional-integral controller.

[0087] 4.2 Constructing the Post-Stage Regulator

[0088] The downstream regulator is a denitrifying agent (NH3) flow regulator. The setpoint is a weighted average of the upstream regulator output and the model control parameters, in kg / h. The feedback value is the actual denitrifying agent feedback value, in the same unit as the setpoint. The regulator output is the denitrifying agent dosing device speed setpoint, in rpm. A proportional-integral controller is used.

[0089] The above cascaded PID controllers are deployed in the basic automation system (Siemens S7-417H).

[0090] The specific steps of further step 5 are as follows:

[0091] 5.1 Real-time parameter acquisition and model parameter input

[0092] The current model input parameters (including the dust collector outlet flue gas volume, dust collector outlet NO content, dust collector outlet NO2 content, direct-fired furnace outlet temperature, direct-fired furnace outlet flue gas pressure, chimney outlet flue gas NO content, chimney outlet flue gas NO2 content, and chimney outlet flue gas NH3 content, a total of 8 parameters) are collected online in real time and transmitted to the neural network model that has been learned from the previous data.

[0093] 5.2 Online Model Execution

[0094] Based on actual operating conditions, the frequency of model operation is adjusted online to minimize the number of model operations while ensuring control accuracy.

[0095] This invention improves the control precision and response speed of denitrifying agent addition; it is highly adaptable and suitable for various industrial application scenarios; and the control effect can be continuously optimized through machine learning models.

Claims

1. A method for improving the accuracy of denitrifying agent dosing and the response speed of the control system, characterized in that... Includes the following steps: S1. Data advance acquisition: Obtain system operating status parameters through sensor system and prepare data training set; S2. Model building: Construct a denitrification agent consumption prediction model. The denitrification agent consumption prediction model is obtained by machine learning training based on neural network algorithm and its model control parameters are initialized. S3. Model learning and parameter generation: After processing the system operation status parameters in step S1, the parameters are fed into the denitrification agent consumption prediction model in step S2 for learning. S4. Controller construction: Construct a cascaded PID controller in the L1 system with the NOx concentration at the chimney outlet as the set value and the frequency of the ammonia water motor as the output value. S5, the recommended dosage of denitrifying agent, serves as a feedforward parameter for the cascade PID controller, acting as a reference calculation data between controller stages. It participates in the L1 closed-loop control to adjust the operating parameters of the ammonia water motor, thereby achieving high-precision control based on model prediction.

2. The method for improving the accuracy of denitrifying agent dosing and the response speed of the control system according to claim 1, characterized in that: In step S1, the collected system operating status parameters include the flue gas volume at the dust collector outlet, the NO content at the dust collector outlet, the NO2 content at the dust collector outlet, the temperature at the direct-fired furnace outlet, the flue gas pressure at the direct-fired furnace outlet, the NO content at the chimney outlet, the NO2 content at the chimney outlet, and the NH3 content at the chimney outlet.

3. The method for improving the accuracy of denitrifying agent dosing and the response speed of the control system according to claim 1, characterized in that: In step S1, the system operating status parameters are collected at locations before the ammonia reactor, providing preliminary calculations for PID setting parameters. The calculated values ​​form the de facto system feedforward.

4. The method for improving the accuracy of denitrifying agent dosing and the response speed of the control system according to claim 1, characterized in that: In step S4, the steps for constructing the cascaded PID controller are as follows: S4.1 Constructing the pre-stage regulator The pre-regulator is a nitrogen oxide regulator, set to the expected nitrogen oxide concentration in mg / Nm³. 3 The feedback value is the actual feedback value of nitrogen oxides, with the same unit as the set value, and a proportional-integral controller is used. S4.2 Constructing the Post-Stage Regulator The downstream regulator is a denitrifying agent flow regulator. The setpoint is the weighted calculated value of the upstream regulator output and the model control parameters, in kg / h. The feedback value is the actual feedback value of the denitrifying agent, in the same unit as the setpoint. The regulator output is the denitrifying agent dosing device speed setpoint, in rpm. It uses a proportional-integral regulator.

Images