Intelligent control of coal-fired boiler and intelligent prediction and control method of flue gas emission thereof

Abstract

The application discloses an intelligent control coal-fired boiler and a flue gas emission method based on intelligent prediction and regulation, which comprises a hearth, water is supplied into the water-cooled wall of the hearth from an inlet pipe; a coal combustion burner capable of heating the hearth; a pulverized coal pipe capable of supplying pulverized coal to the burner; a pulverized coal supply valve suitable for controlling the pulverized coal pipe; and a monitoring device capable of detecting the flow rate of water entering the inlet pipe and making the valve close when the flow rate is lower than a predetermined value; the hearth is connected with a flue, an SCR denitration device is arranged on the flue, and an intelligent regulation system is arranged. The application realizes precise control of ultra-low emission of tail gas pollutants and efficient operation of the denitration system.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of thermal energy and power engineering, belongs to the field of intelligent control technology of coal-fired boilers, and particularly relates to an intelligent control coal-fired boiler and a method for intelligently predicting and regulating flue gas emission based on the same. BACKGROUND

[0002] The traditional design of coal-fired boilers is to measure the individual performance of all burners in multiple configurations by analyzing the combustion products of the burners as a whole, for example, measuring oxygen and / or carbon dioxide in the flue gas passage.

[0003] The design basis is actually wrong because the traditional design does not provide control or measurement of the fuel-air ratio of individual burners under any specific load on the boiler; and in particular, does not include control measurement or calibration of the fuel-air ratio or load changes on the boiler, which is a common problem when the boiler is associated with a power generation system. In the traditional design, there is no separate fuel flow measurement to the individual burners in the configuration. Such measurement is usually made as a whole within the main fuel supply system of the boiler.

[0004] Flue gas generated by coal-fired power plants is an important source of atmospheric NO x X, resulting in an increasing amount of nitrogen oxide emissions. In response to national policy, industrial technology is developed to reduce NO x X emissions. In response to national policy, to achieve NO x X emission standards, to control the amount of ammonia injection in the SCR system to save operating costs, and to improve the efficiency of denitration, the boiler combustion system and the SCR denitration system need to be continuously optimized and intelligently predicted and controlled.

[0005] The operating conditions of the boiler may frequently change according to actual conditions, causing frequent changes in the flue gas parameters at the outlet of the furnace, resulting in the NO x X concentration at the outlet of the furnace being too high when entering the SCR reactor, affecting the subsequent denitration process and also affecting the denitration efficiency. In addition, the detection of boiler operating condition parameters and flue gas parameter signals of the boiler combustion system in the current NO x X emission control system has a lag, and in addition, the SCR reactor reaction process has a large delay. Therefore, for the SCR denitration process which has a complex reaction mechanism, nonlinearity, large delay, and multivariable coupling, the traditional PID control cannot achieve satisfactory control effect, so the DMC-PID cascade controller combined with the BP neural network can overcome the above difficulties.

[0006] Therefore, it is necessary to develop an intelligent prediction and regulation method for coal-fired flue gas NO xThe emission control method is imperative, and is of great significance for safe and economic operation of a boiler combustion system and an SCR denitration device. SUMMARY

[0007] The application designs an intelligent control coal-fired boiler and an emission control method for coal-fired flue gas NO x based on intelligent prediction and regulation. x The emission control method mainly optimizes the design of a NO x concentration prediction control system in an SCR denitration process, realizes timely and accurate prediction control of ammonia injection amount in the SCR denitration system, and achieves the national NO x ultra-low emission standard with optimal operation cost and denitration efficiency. x The method mainly predicts the NO x concentration at the furnace outlet (the NO x concentration at the SCR reactor inlet) through a BP neural network model, and then maps and approximates the nonlinear relationship between the predicted NO x concentration at the SCR reactor outlet and the input variables online, and feeds back the deviation between the measured and set values of the NO x concentration at the SCR reactor outlet to the DMC prediction controller and the PID controller to accurately and quickly control the ammonia injection amount.

[0008] An intelligent control coal-fired boiler comprises a furnace, water is supplied from an inlet pipe into a water wall of the furnace, a coal combustion burner capable of heating the furnace, a coal powder pipe capable of supplying coal powder to the burner, a coal powder supply valve adapted to control the coal powder supply through the coal powder pipe, and a monitoring device capable of detecting the flow rate of water entering the inlet pipe and causing the valve to be closed when the flow rate is lower than a predetermined value, a furnace is connected to a flue, and an SCR denitration device is arranged on the flue.

[0009] A thermostat is assembled to the outlet of the water wall, and if the outlet water temperature exceeds a predetermined maximum value, the thermostat is operable to cause the coal powder supply valve to be closed.

[0010] An emission control method for coal-fired flue gas NO x based on intelligent prediction and regulation comprises the following steps.

[0011] Step S1: preparing an SCR denitration system, a NO x set value device, a NO xOnline concentration measurement instruments, DMC predictive controllers, and PID controllers;

[0012] Step S2: Collect boiler operating data and perform correlation preprocessing; construct a BP neural network intelligent prediction model to predict the NO at the furnace outlet. x concentration;

[0013] Step S3: Using DMC as the main controller and PID as the secondary controller, a DMC-PID cascade feedback control structure is formed. Rolling optimization and feedback correction are performed, and a BP neural network model is combined to compensate for prediction deviations, thus online mapping and approximating the NO outlet of the SCR denitrification system. x The complex nonlinear relationship between the concentration prediction value and various variables allows for precise pre-control of the ammonia injection amount in the SCR denitrification system based on different operating conditions of the boiler combustion system, thus avoiding delays in the denitrification system.

[0014] In step S1, the NO at the outlet of the SCR denitrification system is... x The concentration setting of the setter is set to be lower than the ultra-low emission standard (50 mg / m³). 3 Slightly lower.

[0015] In step S1, NO is installed at the outlet of the SCR denitrification system. x Online concentration measurement instruments can monitor and measure NO at the outlet of SCR denitrification systems in real time. x The concentration is then used as a feedback input signal to the BP neural network and the DMC predictive controller.

[0016] In step S1, operating data of the boiler combustion system and the denitrification operation data of the SCR denitrification system are collected under different operating conditions at different time periods. It is necessary to include data under different operating conditions to increase the diversity of data, which is beneficial to the prediction effect of the BP neural network model under different operating conditions.

[0017] The collected operational data needs to be preprocessed, including normalization and correlation calculation. Then, the BP neural network is trained and tested to make the BP neural network model more stable and reduce the error of the predicted objective function.

[0018] Based on the collected historical data of boiler combustion system and SCR reactor denitrification operation, a BP neural network prediction model was constructed, consisting of an input layer, a hidden layer, and an output layer. Using the constructed BP neural network, the NO inlet of the SCR denitrification system was predicted in real time. x Concentration and NO at the outlet of the SCR denitrification system x concentration.

[0019] According to the historical data of the boiler combustion system and the SCR reactor denitration operation, the input variables related to the objective function are determined, and the number of hidden layer neurons and the number of hidden layers are established according to the number of input variables, which relates to the accuracy of the BP neural network prediction result, and the NOx concentration at the furnace outlet can be predicted x The concentration of the SCR denitration system outlet position NO x The complex nonlinear relationship between the predicted value and each variable.

[0020] In the step S1, the PID controller inputs the NOx concentration signal of the DMC prediction controller, and can calculate and control the ammonia injection amount required by the SCR denitration system in real time, so as to achieve preliminary control of the ammonia injection valve opening degree. x The concentration of the SCR denitration system outlet position NO x The complex nonlinear relationship between the predicted value and each variable. x The complex nonlinear relationship between the predicted value and each variable. x The complex nonlinear relationship between the predicted value and each variable. x The complex nonlinear relationship between the predicted value and each variable. x The complex nonlinear relationship between the predicted value and each variable.

[0021] In order to facilitate further understanding of the technical scheme, in more detail, the technical scheme design idea of the application is implemented as follows:

[0022] First, the whole prediction control system is designed: a BP neural network model is constructed to accurately and timely predict the NOx concentration at the furnace outlet x The concentration of the SCR reactor, and the DMC-PID cascade controller preliminarily controls the ammonia injection amount required by the SCR reactor according to the predicted value of the NOx concentration at the furnace outlet x The complex nonlinear relationship between the predicted value and each variable. x The complex nonlinear relationship between the predicted value and each variable. x The complex nonlinear relationship between the predicted value and each variable. x The complex nonlinear relationship between the predicted value and each variable. x The complex nonlinear relationship between the predicted value and each variable.

[0023] The specific operation steps are as follows:

[0024] Step 1: Based on the historical operation data of the coal-fired power plant combustion system, a plurality of boiler operation parameters are collected, and a BP neural network model for accurately predicting the NOx concentration at the furnace outlet is established. x The concentration of the SCR reactor, and the DMC-PID cascade controller preliminarily controls the ammonia injection amount required by the SCR reactor according to the predicted value of the NOx concentration at the furnace outlet

[0025] Note: The BP neural network model includes an input layer, a hidden layer and an output layer, and the working principle is as follows:

[0026] 1. This design has many input variables. To briefly explain the principle, it is assumed that the input layer has three variables (boiler parameters); the hidden layer is one layer containing two neurons; and the output layer is a single objective function (furnace outlet NO). x concentration);

[0027] 2. First, the hidden layer connecting the input and output layers has an activation function that processes the data, typically the sigmoid function:

[0028] 3. Input data for the input layer: Where x1, x2, and x3 represent boiler operating parameters, the BP neural network will simulate the weight matrices of the two neurons in the hidden layer corresponding to x1, x2, and x3: The hidden layer's corresponding output layer weight matrix: w0 = [w7 w8], and the hidden layer neuron threshold matrix: Output layer threshold: b0;

[0029] 4. Based on the above formula,

[0030] Substitute the Z value into the sigmoid function: The values ​​after hidden layer activation are obtained: Final objective function prediction (furnace outlet NO) x Concentration): y = w0·h + b0 = w7h1 + w8h2 + b0;

[0031] The sigmoid function is the activation function for neurons in the hidden layer. Each parameter in the input layer corresponds to a threshold and a weight for each neuron in the hidden layer, as shown in the above formula. and Then, the independent variable Z of the sigmoid function is calculated by inputting the variables, corresponding thresholds, and weights. The Z value is then substituted into the f(z) function, and the input layer variables are used. The data after activation by the hidden layer activation function is Each neuron in the hidden layer also has a corresponding threshold and weight for the target function y (NOx concentration at the furnace outlet), as shown in the above formulas [w7 w8] and b0. The target function y (NOx concentration at the furnace outlet) is calculated from the data h after the activation function and the threshold and weight. x Concentration), that is, the above formula y=w0·h+b0=w7 h1+w8 h2+b0; through this calculation step, the target function is predicted by the three layers of input layer, hidden layer and output layer, mainly by the threshold and weight associated between the data of each layer and the data of the next layer, and then calculated by the activation function.

[0032] 5. The above steps outline the entire principle and formula for BP neural network prediction. The predicted value is the result predicted by the BP neural network. The weights and thresholds between each layer of data are also values ​​where the BP neural network simulates the minimum error. In practice, it is not necessary to write out these formulas because each boiler operating parameter and furnace outlet NO... x The predicted concentration of NO will be converted into weights and thresholds provided by a BP neural network. Furthermore, how do boiler operating parameters affect NO at the furnace outlet? x The concentration, and which factors are the most important influencing factors, will be demonstrated in the correlation calculations in later steps.

[0033] Using a BP neural network model, simulation training and prediction are performed on the unit's operating data under different current operating conditions to predict the furnace outlet NO under the current boiler operating conditions. x concentration.

[0034] Step 2: Furnace outlet NO in Step 1 x The predicted concentration value is the NO at the SCR reactor inlet. x Based on the current nitrogen oxide concentration, the required ammonia injection rate for the system under the current operating conditions is calculated. The signal for the required ammonia injection rate is transmitted to the DMC-PID controller, which then initially controls the ammonia injection valves required by the SCR system and measures the NO concentration at the SCR reactor outlet. x The true value.

[0035] Step 3: Use a BP neural network model to approximate the outlet NO of the SCR reactor online. x Predicted concentration values ​​and flue gas parameters, furnace outlet NO x Predicted concentration, ammonia injection valve opening, and SCR reactor outlet NO x The complex nonlinear relationship between the measured values.

[0036] Note: The BP neural network here serves the same function as in step 1, which is to reduce the NO at the SCR reactor outlet. x The measured values ​​are used as the predicted values ​​of the objective function, which can reflect the relationship between these important influencing factors and the NO at the SCR reactor outlet. x The nonlinear relationship between the measured values ​​is explained by the BP neural network, which provides the weights and thresholds between the data in each layer. The specific principle is explained in step 1.

[0037] This signal, along with the deviation feedback signal, is transmitted to the DMC-PID cascade controller. Based on this signal, the DMC-PID cascade controller readjusts the ammonia injection valve to reduce the NO at the SCR reactor outlet. x The concentration meets the emission standards.

[0038] At this point, all the process design steps have been completed, based on systematic intelligent prediction and full-process control of NO in coal-fired flue gas. x The emission control system has been basically implemented. The ammonia injection rate in the SCR reactor is controlled by a combination of feedforward and deviation feedback signals, thereby reducing the NO content in the flue gas. x Emissions meet national standards, and can even reach ultra-low emissions, while also ensuring the economical operation of all equipment.

[0039] The establishment of the BP neural network prediction model in step 1 should be based on the actual normal operation of the power plant unit and the selection of boiler operating parameters.

[0040] Furthermore, based on extensive literature review and calculation of the correlation between parameters, a correlation of greater than 0.5 was selected between input and output variables. Correlation, as the name suggests, refers to the connection and mutual influence between data. Some data exhibit a promoting relationship, known as positive correlation (value greater than 0). The larger the value, the stronger the connection and mutual influence, and these can be considered main factors or important influencing factors. Other data exhibit an inhibiting relationship, known as negative correlation (value less than 0). The smaller the value, the closer the connection and the stronger the influence. In typical boiler equipment operation, a correlation of around 0.5 indicates a relatively close connection and strong mutual influence between the data. The output variable is the furnace outlet NO mentioned above. x Concentration (NO at the inlet of the SCR reactor) x concentration).

[0041] Furthermore, the selected boiler operating parameters are used as input variables: boiler load, primary air volume, secondary air volume, burnout air volume, oxygen content, furnace temperature, etc., which are used by the neural network to predict the NO at the furnace outlet. x Concentration is the input variable.

[0042] Furthermore, based on the selected input parameters, a BP neural network model is established. Due to the large number of input variables, and considering the model's predictive performance, the hidden layer of the neural network is designed as a double layer.

[0043] Furthermore, the selected input variable data needs to be preprocessed. First, abnormal data in each group, including data under extreme working conditions, should be removed. Then, similar data, such as multiple groups of data under the same or similar working conditions, should be eliminated.

[0044] Furthermore, the BP neural network model predicts the NO at the furnace outlet. x The concentration is the NO at the SCR reactor inlet. x Concentration, based on NO xThe concentration is calculated to determine the corresponding ammonia injection rate. The DMC-PID cascade controller controls the opening of the ammonia injection valve according to the required ammonia injection rate of the SCR reactor, thereby initially and accurately controlling the ammonia injection rate in the SCR reactor.

[0045] Furthermore, the NO at the SCR reactor outlet is approximated using an online neural network mapping. x The nonlinear relationship between the concentration prediction and variable parameters, along with the deviation feedback signal, is transmitted to the DMC predictive controller to promptly adjust and correct the required ammonia injection rate to meet the NO emission requirements of the SCR denitrification system. x Concentration setpoint requirements must comply with national NO standards. x Emission requirements and standards. When the predictive control system detects changes in the operating conditions of the boiler combustion system and SCR reactor, the system adaptively adjusts the DMC-PID parameters to achieve optimized control.

[0046] The present invention has the following advantages:

[0047] (1) The system can directly predict and control the required ammonia injection amount of the SCR denitrification system based on the boiler operating parameters under different operating conditions in real time. This allows the system to be adjusted in advance under normal economic operation conditions. By directly controlling the source, it avoids the delay and lag caused by the process adjustment based on the results of general control systems. It also overcomes the control problems caused by the delay of the detection system and the large delay characteristics of the SCR denitrification system. This provides more reaction time for the control system, reduces control overshoot, and improves the system's fault tolerance. While saving ammonia injection amount, it also reduces ammonia slip and achieves accurate control of ammonia injection amount. This meets both national emission standards and the economic operation of the system.

[0048] (2) This control system is a highly automated control system, which greatly reduces the operation of staff, thereby reducing human error and making the ammonia injection control of the SCR denitrification system more precise.

[0049] (3) The deviation between the nitrogen oxide concentration at the outlet of the SCR denitrification system and the set value can be reduced to an acceptable range. The required ammonia injection amount can be corrected according to different deviation signals to achieve NO under different operating conditions of the system. x Stable emissions.

[0050] (4) NO from coal-fired flue gas under full-process control x The emission control system has a high system response rate, low investment, short renovation period, and high denitrification efficiency. Attached Figure Description

[0051] Figure 1 This is the basic structure of a single-layer hidden-layer BP neural network.

[0052] Figure 2The basic structure of the double-layer implicit layer BP neural network.

[0053] Figure 3 The control flow chart of the required ammonia amount of the system.

[0054] Figure 4 The neural network prediction of the furnace outlet NO x Concentration flow chart.

[0055] Figure 5 The SCR denitration system NO x Emission prediction control overall flow chart. DETAILED DESCRIPTION

[0056] In order to make the content of the present application more easily understood, the present application will be further described in detail below according to specific implementation cases and in conjunction with the accompanying drawings. It should be understood that these examples are only used to illustrate the present application and are not used to limit the scope of the present application. After reading the present application, those skilled in the art can make various equivalent modifications of the present application, which all fall within the scope defined by the appended claims.

[0057] First of all, it needs to be pointed out that:

[0058] The boiler combustion system refers to the combination of equipment and corresponding flue gas, air, and coal (pulverized coal) pipelines required for the full combustion of fuel in the furnace of the boiler and the discharge of flue gas generated by combustion into the atmosphere. The operating condition is directly related to the NO x Concentration at the furnace outlet. The SCR denitration system is a complex control system, and the reaction is affected by factors such as catalyst activity, flue gas temperature, flue gas flow rate, and oxygen concentration. The mathematical model established according to the mechanism method can still work well when the working condition is unchanged, but the mathematical model will change greatly when the working condition changes. Therefore, in the face of ultra-low emission control of nitrogen oxides, the boiler combustion system and the SCR denitration system need to be continuously optimized and predicted controlled.

[0059] Figure 1 The single-layer implicit layer BP neural network structure includes an input layer, an implicit layer, and an output layer. The output layer is the target function to be predicted. The number of implicit layer neurons is an empirical formula: h is the number of implicit layer neurons, m refers to the number of input layer nodes, that is, the dimension of the input variable, n refers to the number of output variables, that is, the number of target functions, and c is a number between 0 and 10.

[0060] Figure 2 The double-layer implicit layer BP neural network model. Due to the increase in the number of input variables, the number of implicit layers will be increased to improve the accuracy of the model. This design adopts a double-layer implicit layer, and the remaining design rules are consistent with those of the single-layer implicit layer BP neural network.

[0061] Figure 3 is calculated according to the flue gas parameters and the SCR reactor inlet NO x concentration, and the signal is transmitted to the PID controller for opening of the ammonia injection valve.

[0062] Figure 4 The BP neural network predicts the furnace outlet NO x concentration flow chart, and the steps include: data collection, parameter selection, data preprocessing, BP neural network construction, data simulation training and prediction.

[0063] Preferably, the data preprocessing content is as follows: firstly, the correlation between the boiler operation parameters and the furnace outlet NO x concentration is calculated by using the Correl function or the Pearson function of the Excel software, the boiler operation parameters with the correlation greater than 0.5 are selected as the main factors and also as the input variables of the BP neural network model; then, the data of each group of input variables are processed, for example, the data not conforming to the actual situation under extreme working conditions are removed, the data should include the data under various working conditions as much as possible, and finally the BP neural network structure is established.

[0064] Figure 5 The DMC-PID control unit predicts the SCR denitration system NO x emission, the prediction includes the BP neural network prediction of the furnace outlet NO x concentration (the SCR reactor inlet NO x concentration), and the SCR reactor outlet NO x concentration prediction, the SCR reactor outlet NO x concentration prediction value and the SCR reactor outlet NO x concentration measured value and the set value are transmitted to the DMC prediction controller, and the signal is accepted by the PID controller for optimized control.

[0065] The embodiment is based on the specific implementation of the SCR denitration system of a 300 MW coal-fired power plant front and back wall offset boiler, the coal-fired flue gas NO x emission control system based on systematic intelligent prediction and whole-process regulation is designed and researched, and the specific implementation process is as follows:

[0066] The present application mainly proposes a coal-fired flue gas NO x emission control method based on systematic intelligent prediction and whole-process regulation, mainly relates to the intelligent prediction control of the flue gas NO x emission by combining the BP neural network model prediction and the DMC-PID control unit.

[0067] The first step in this invention is to collect a large amount of historical boiler operating data. This involves referring to a large number of documents, boiler operating principles, denitrification expertise, and industry experience to select certain boiler operating parameters, including primary air volume, secondary air volume, burnout air volume, oxygen content, coal mill combination method, boiler load, furnace temperature, and other boiler operating parameters.

[0068] Based on the parameters selected above, the boiler operating parameters are set as input variables, and the furnace outlet NO... x Concentration is set as the output variable. The correlation between the input and output variables is calculated, and the input variables with a correlation greater than 0.5 are selected. Then, a two-layer hidden layer neural network model is constructed based on the obtained input and output variables.

[0069] It is worth noting that when using neural networks to simulate data, the data should be normalized and denormalized to bring it all within the range of [-1, 1]. This can reduce the errors caused by numerical problems.

[0070] Normalization formula: y = (y max -y min )*(xx min ) / (x max -x min )+y min ; where y max y is 1 min x is 1 max and x min To collect the maximum and minimum values ​​of each row or column of each variable in the data, y is the normalized data.

[0071] Based on boiler operating parameters under different operating conditions, a constructed neural network model is used to predict NO at the furnace outlet. x Concentration. Then it needs to be based on NO. x The required amount of liquid ammonia, the reducing agent, for the SCR reactor is calculated. This ammonia quantity signal is then transmitted to the DMC-PID control unit for initial and rapid control of the ammonia injection rate into the SCR reactor. Simultaneously, the NO concentration at the SCR reactor outlet is measured. x concentration.

[0072] The required ammonia injection rate is based on the NO inlet temperature of the SCR reactor. x Concentration calculation, using predicted NO at the furnace outlet x The required ammonia injection rate can be calculated by multiplying the concentration by the flue gas volume, and then by the ammonia-nitrogen molar ratio. Here, "flue gas volume" refers to the actual measured volume of boiler flue gas, measured in Nm³. 3"H2N mole ratio" refers to the ratio of the concentration of ammonia and nitrogen oxide in the denitration, which is a common concept in denitration, and is generally 0.7-0.9. Therefore, the calculation formula of "H2N demand" is: H2N demand = SCR reactor inlet nitrogen oxide x flue gas volume x H2N mole ratio.

[0073] As a further improvement, as the predicted NO x concentration increases, the H2N demand increases nonlinearly; as the flue gas volume increases, the H2N demand increases nonlinearly.

[0074] Further improvement, as the predicted NO x concentration increases, the H2N demand increases at an increasingly large rate; as the flue gas volume increases, the H2N demand increases at an increasingly large rate.

[0075] As a preferred embodiment, the H2N demand control method is as follows:

[0076] The controller stores standard data H2N demand R, predicted NO x concentration C, and flue gas volume V, and when the predicted NO x concentration C and the flue gas volume V are met, the H2N demand R meets the requirements.

[0077] The standard data represents data that meets certain flue gas emission conditions. For example, it can be a requirement to meet the NO x concentration in the flue gas within a certain range.

[0078] Then, when the NO x concentration becomes c and the flue gas volume becomes v, the H2N demand r meets the following requirements:

[0079] (v*c) / (V*C)=a*Ln((r-R 标准 ) / (R-R 标准 ))+b, a, b are parameters, and satisfy the following formula:

[0080] (r-R 标准 ) / (R-R 标准 )<1, 1.041<a<1.044, 1.0<b<1.005;

[0081] (r-R 标准 ) / (R-R 标准 )=1, b=1;

[0082] (r-R 标准 ) / (R-R 标准 )>1, 1.045<a<1.052; 0.991<b<1;

[0083] As a preferred embodiment, (r-R 标准) / (R-R 标准 )<1, a=1.042, b=1.003.

[0084] As a preference, (r-R 标准 ) / (R-R 标准 )<1, a=1.048, b=0.997.

[0085] As a preference, (r-R 标准 ) / (R-R 标准 )<1, as (r-R 标准 ) / (R-R 标准 ) increases, a becomes larger and b becomes smaller.

[0086] As a preference, (r-R 标准 ) / (R-R 标准 )>1, as r / R increases, a becomes larger and b becomes smaller. This design makes the flue gas emission meet the requirements as soon as possible.

[0087] Where R 标准 is the NO x concentration value in the flue gas to meet the flue gas emission requirements for normal operation of the boiler, which can preferably be the upper limit value of the required NO x concentration.

[0088] In the formula of the above mode, the following conditions need to be met: 0.88<=(r-R 标准 ) / (R-R 标准 )<=1.13.

[0089] As a preference, the controller stores multiple sets of standard data.

[0090] As a preference, when multiple sets of standard data are met, the set of v and c with the minimum value of ((1-v / V) 2 +(1-c / C) 2 ) is selected; of course, the first set of v and c that meets the requirements can also be selected, or a set of v and c that meets the conditions can be randomly selected.

[0091] Through the design of the operation mode of the above intelligent control, compared with the linear input of the amount of ammonia gas, the amount of input ammonia gas can be more accurately calculated, the boiler tail gas emission can meet the emission requirements faster and better, and the system intelligence is further improved, meeting the requirements of energy saving and environmental protection.

[0092] Then, according to the neural network model, the nonlinear relationship between the predicted value of the NO x concentration at the outlet of the SCR reactor and the variable parameters is mapped and approximated online, and it is compared with the NO xThe deviation signals of the concentrations are jointly transmitted to a DMC-PID control unit, which reversely controls and adjusts the ammonia injection amount required by the SCR system, and timely and accurately regulates the ammonia injection amount in the SCR system.

[0093] The feedback signals are transmitted to a DMC-PID cascade controller, wherein the parameters of the control unit need to be set to obtain optimal parameters according to the field conditions, and the steps are as follows:

[0094] (1) During initial debugging, the characteristic parameters of the system under different conditions are obtained according to field tests, an input-output prediction control system model of the SCR denitration system is established, and the PID parameters can be determined through the model.

[0095] (2) In the running process, the boiler operation data and the SCR reactor working data are collected in real time, and the model is corrected, and then based on the corrected model, the optimal PID parameters under the current condition are obtained by using an optimization algorithm, so as to realize self-adaptive PID control.

[0096] Through the above-mentioned neural network combined with the DMC-PID cascade controller control system, the national NO x emission standard can be reached, so as to reduce the operation cost and improve the denitration efficiency. Because the present application simultaneously involves prediction and control, and selects the DMC-PID cascade controller, the system response rate will be faster in the practical process, and the ammonia control amount of the rear end can be directly reflected according to the different conditions of the front end boiler combustion system, so that the reaction is more timely, in addition, the SCR outlet NO x concentration can be controlled in a lower range, and the deviation from the set value can be smaller, so that the NO x emission concentration is lower, and the accuracy of the control system is improved.

[0097] In order to ensure that the algorithm can be applied in the field, and to protect the control algorithm, the implementation of the present scheme is based on the optimization control NO x emission platform.

[0098] The present applicant has referred to a large number of literatures on the application of BP neural network to the prediction function of thermal power plants, and found that the BP neural network has good nonlinear approximation ability and stronger robustness when facing random sudden changes of unit operation conditions and uncertain nonlinear relationships in the actual system, and the prediction performance is more stable, while the dynamic matrix control (DMC) has the disadvantage of being difficult to timely predict control, thereby causing prediction deviation. Therefore, the applicant uses the BP neural network to compensate the DMC prediction error online, connects the BP neural network and the DMC prediction controller in series, and further, the BP neural network can online approximate the SCR denitration system outlet NO xThe complex nonlinear relationship between the concentration prediction value and each variable minimizes the prediction error and realizes timely and accurate control of the ammonia injection amount in the SCR denitration system.

[0099] In the present scheme, the collection of operation data is performed in a time period of device operation to include as much operation data as possible under all working conditions, without manual collection, thereby reducing the error of manual collection. The big data collection can also increase the number of collection points or the range of collection as much as possible. The diversity of data is beneficial to the fitting and prediction performance of the BP neural network model, so that the error of the objective function is smaller, and the model prediction result is more stable.

[0100] In addition, the establishment of the BP neural network model mainly involves the establishment of the hidden layer. Generally, the number of layers of the hidden layer is one or two, which is determined according to the dimension of the input variable or the comparison of the prediction result.

[0101] The formula for determining the number of neurons in the hidden layer of the BP neural network is h is the number of neurons, m is the number of input layer nodes, that is, the dimension of the input variable, n is the number of output variables, that is, the number of objective functions, and c is a number between 0 and 10. The number of neurons in the hidden layer of different models is not the same, which needs to be determined according to the comparison of the prediction fitting effect of the actual situation.

[0102] Regarding the determination of the parameters of the DMC-PID cascade controller, the initial parameters of the control unit need to be obtained by debugging based on the operation data of multiple boiler systems and denitration systems, then tested according to the field operation data, and the model is corrected, then based on the corrected model, the optimal DMC-PID parameters under the current working condition are obtained by using the optimization algorithm, so as to realize the adaptive DMC-PID control.

[0103] The above specific implementation methods are only described according to the technical scheme of the present application, and are used as examples, but the protection scope of the present application is not limited thereto. The technical personnel in the art can modify or replace the technical scheme of the present application, but all should be included in the protection scope of the present application, and the protection scope of the present application should be subject to the description in the claims.

Claims

1. A method for intelligent predictive regulation of NO in coal-fired flue gas x The emission control method is characterized by, Specifically, the following steps are included: Step S1: Prepare the SCR denitrification system and NO x Setter, NO x Online concentration measurement instruments, DMC predictive controllers, and PID controllers; Step S2: Collect boiler operating data and perform correlation preprocessing; construct a BP neural network intelligent prediction model to predict the NO at the furnace outlet. x Concentration and NO at SCR denitrification outlet x concentration; In step S2, boiler combustion system operation data and SCR denitrification system operation data are collected under different operating conditions at different time periods. It is necessary to include data under different operating conditions to increase the diversity of data, which is beneficial to the prediction effect of the BP neural network model under different operating conditions. The BP neural network model is used to simulate and train the operating data under different current operating conditions to predict the NO at the furnace outlet under the current operating conditions. x Concentration and NOx concentration at the SCR denitrification outlet; Step S3: Using DMC as the main controller and PID as the secondary controller, a DMC-PID cascade feedback control structure is formed. Rolling optimization and feedback correction are performed, and a BP neural network model is combined to compensate for prediction deviations, thus online mapping and approximating the NO outlet of the SCR denitrification system. x The complex nonlinear relationship between the concentration prediction value and various variables allows for precise pre-control of the ammonia injection amount in the SCR denitrification system based on different operating conditions of the boiler combustion system, thus avoiding delays in the denitrification system. In step S3, the PID controller predicts the NO input from the DMC controller. x The concentration signal is used to calculate and control the amount of ammonia required by the SCR denitrification system in real time, thereby achieving preliminary control of the ammonia injection valve opening. Then, combined with the BP neural network model, prediction deviation compensation, rolling optimization and feedback correction are performed to quickly and accurately control the ammonia injection valve opening, which can avoid the delay and lag disadvantages brought about by general control systems. Furnace outlet NO x The predicted concentration value is the NO at the SCR reactor inlet. x Based on the current nitrogen oxide concentration, the required ammonia injection rate for the system under current operating conditions is calculated. This required ammonia injection rate signal is transmitted to the DMC-PID controller. The DMC-PID controller controls the ammonia injection valve opening according to the required ammonia injection rate for the SCR reactor, thereby initially controlling the ammonia injection rate in the SCR reactor and measuring the NO₂ at the SCR reactor outlet. x The true value; Online mapping of NO at the SCR reactor outlet using a BP neural network model x Predicted concentration values ​​and flue gas parameters, furnace outlet NO x Predicted concentration, ammonia injection valve opening, and SCR reactor outlet NO x The nonlinear relationship between the measured values, along with the deviation feedback signal, is transmitted to the DMC predictive controller to adjust and correct the required ammonia injection rate in a timely manner, thereby meeting the NO emission requirements at the SCR denitrification system outlet. x Concentration setpoint requirements must comply with national NO standards. x Emission requirements standards; when the predictive control system detects changes in the operating conditions of the boiler combustion system and SCR reactor, the system will adaptively adjust the DMC-PID parameters to achieve optimized control.

2. The method for intelligent predictive regulation of NO in coal-fired flue gas according to claim 1 x The emission control method is characterized by, In step S1, the NO at the outlet of the SCR denitrification system is... x The concentration setting of the setter is 50 mg / m³. 3 Slightly lower.

3. The method for intelligent predictive regulation of NO in coal-fired flue gas according to claim 1 x The emission control method is characterized by, In step S1, NO is installed at the outlet of the SCR denitrification system. x Online concentration measurement instruments can monitor and measure NO at the outlet of SCR denitrification systems in real time. x The concentration is then used as a feedback input signal to the BP neural network and the DMC predictive controller.

4. The method for intelligent predictive regulation of NO in coal-fired flue gas according to claim 1 x The emission control method is characterized by, The collected operational data needs to be preprocessed, including normalization and correlation calculation. Then, the BP neural network is trained and tested to make the BP neural network model more stable and reduce the error of the predicted objective function.

5. The method for intelligent predictive regulation of NO in coal-fired flue gas according to claim 1 x The emission control method is characterized by, Based on the collected historical data of boiler combustion system and SCR reactor denitrification operation, a BP neural network prediction model was constructed, consisting of an input layer, a hidden layer, and an output layer. Using the constructed BP neural network, the NO inlet of the SCR denitrification system was predicted in real time. x Concentration and NO at the outlet of the SCR denitrification system x concentration.

6. The method for intelligent predictive regulation of NO in coal-fired flue gas according to claim 5 x The emission control method is characterized by, Based on historical data from the boiler combustion system and SCR reactor denitrification operation, input variables with high correlation to the objective function are identified. The number of hidden layer neurons and the number of hidden layers are then determined according to the number of input variables. This is crucial to the accuracy of the BP neural network prediction results. This step can achieve NO reduction at the furnace outlet. x Concentration prediction and NO concentration at the outlet of the SCR denitrification system x The concentration prediction values ​​have a complex nonlinear relationship with various variables.

Images