Feedforward Control Method for Flue Gas Denitration System Based on Signal Reconstruction

Abstract

The present invention relates to the technical field of flue gas denitration for coal-fired units, and specifically provides a feedforward control method for a flue gas denitration system based on signal reconstruction, which mainly includes three aspects: SCR inlet NOx signal reconstruction, flue gas flow signal reconstruction, and formation of a feedforward control command for ammonia injection amount. It can effectively overcome the problem of poor control accuracy of ammonia injection amount caused by measurement signal errors at the current stage, effectively achieve precise feedforward control of ammonia injection amount, avoid large fluctuations in the NOx concentration at the SCR outlet, save ammonia injection amount while meeting environmental protection indicators; the feedforward control logic uses the reconstructed SCR inlet NOx signal, effectively solving the time delay problem existing in the current inlet NOx measurement signal. When the unit load changes, the inlet NOx concentration parameter can be obtained in a timely manner, overcoming the influence of measurement time delay on the control system; the feedforward control logic uses the reconstructed flue gas flow, and can obtain an accurate flue gas flow signal online, making the control value of ammonia injection amount meet the index requirements of ammonia-nitrogen molar ratio.

Description

Technical Field

[0001] The present invention relates to the technical field of flue gas denitration for coal-fired units, and particularly to a feedforward control method for a flue gas denitration system based on signal reconstruction. Background Art

[0002] Coal-fired power plants generally adopt a catalyst reduction system (SCR system) installed at the tail position of the flue duct. By injecting a reducing agent NH3 into the flue gas, NH3 and NOx are fully mixed in the flue duct, and then under the action of the reducing agent, NOx is converted into nitrogen and water. The key to this method is to accurately control the molar ratio of NH3 and NOx.

[0003] The SCR inlet NOx concentration is one of the key parameters for calculating the molar ratio of NH3 and NOx. At present, the measurement of the SCR inlet NOx concentration needs to be collected and analyzed through a continuous emission monitoring system for flue gas (CEMS) to obtain the measured value. This results in a large time delay in the CEMS measurement data. When the NOx concentration fluctuates greatly and at a relatively fast speed, it will seriously affect the removal effect of NOx.

[0004] In addition, the flue gas flow rate is another key parameter for calculating the molar ratio of NH3 and NOx. At present, it is difficult to have a large-scale flue gas flow rate measuring device that meets the control requirements and is relatively accurate. Especially after the SCR denitration transformation, the boiler flue duct layout is irregular and cannot meet the requirements of general flow measuring devices for pipelines. The inaccuracy of the flue gas flow rate measurement data will cause the automatic control effect of the denitration control system to be difficult to meet the performance requirements.

[0005] Based on this, for the SCR denitration system of coal-fired units, a scientific and practical signal reconstruction method for the SCR inlet NOx concentration and flue gas flow rate is urgently needed to achieve accurate feedforward control of the ammonia injection amount and effectively control the NOx emissions of coal-fired units. Summary of the Invention

[0006] In view of the above problems, an embodiment of the present invention provides a feedforward control method for a flue gas denitration system based on signal reconstruction.

[0007] On the one hand, an embodiment of the present invention provides a feedforward control method for a flue gas denitration system based on signal reconstruction, and the method includes:

[0008] Calculating the delay time of the SCR inlet NOx concentration measurement signal;

[0009] Determining the initial characteristic variables affecting the SCR inlet NOx concentration, and applying the mutual information method to determine the pure time delay between the initial characteristic variables and the SCR inlet NOx concentration;

[0010] Add pure delay to the initial characteristic variables to obtain intermediate variables, and use the maximum correlation and minimum redundancy method to screen out characteristic variables from the intermediate variables;

[0011] Perform delay processing on the characteristic variables using the delay time and pure delay to obtain input variables;

[0012] Send the input variables into the pre-trained NARX neural network to obtain the reconstructed SCR inlet NOx concentration;

[0013] Obtain the measured value of the total air volume, and use the functional relationship between the measured value of the total air volume and the flue gas flow rate to obtain the reconstructed flue gas flow rate;

[0014] According to the NOx removal principle, use the reconstructed SCR inlet NOx concentration and flue gas flow rate to obtain the ammonia injection amount required for the feedforward control of the flue gas denitration system.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention includes three aspects: SCR inlet NOx signal reconstruction, flue gas flow rate signal reconstruction, and formation of the feedforward control instruction for ammonia injection amount. It can effectively overcome the problem of poor control accuracy of ammonia injection amount caused by measurement signal errors at the present stage, effectively realize the accurate feedforward control of ammonia injection amount, avoid large fluctuations in the SCR outlet NOx concentration, save ammonia injection amount while meeting environmental protection indicators; The feedforward control logic of the present invention uses the reconstructed SCR inlet NOx signal, effectively solves the time delay problem existing in the inlet NOx measurement signal at the present stage. When the unit load changes, the inlet NOx concentration parameter can be obtained in a timely manner, overcoming the influence of measurement time delay on the control system; The feedforward control logic of the present invention uses the reconstructed flue gas flow rate. During the operation of the unit, an accurate flue gas flow rate signal can be obtained online, so that the control value of the ammonia injection amount meets the index requirements of the ammonia-nitrogen molar ratio, avoiding large fluctuations in the SCR outlet NOx caused by too much or too little ammonia injection amount.

[0016] Optionally, the process of calculating the delay time of the SCR inlet NOx concentration measurement signal includes:

[0017] Obtain the oxygen content data in the flue gas within a predetermined time period;

[0018] Obtain the SCR inlet NOx concentration data within a predetermined time period;

[0019] Embed different pure delay times into the obtained oxygen content data to obtain the reconstructed oxygen content data;

[0020] Calculate the person correlation coefficient between the reconstructed oxygen content data and the obtained SCR inlet NOx concentration data respectively. The calculation formula is as follows:

[0021]

[0022] Among them, X represents the time series of the oxygen content data after reconstruction, that is, O2(t - t d ), where t d represents the embedded pure delay time; Y represents the time series of the SCR inlet NOx concentration data, that is, NO x (t); cov(X, Y) represents the covariance of the sequences X and Y; δ X represents the standard deviation of the sequence X, and δ Y represents the standard deviation of the sequence Y;

[0023] When the correlation coefficient ρ(X, Y) is the largest, the pure delay time t d embedded in the reconstructed oxygen content data O2(t - t d ) is used as the delay time of the SCR inlet NOx concentration measurement signal.

[0024] Optionally, to determine the initial characteristic variables affecting the SCR inlet NOx concentration, the process of applying the mutual information method to determine the pure delay between the initial characteristic variables and the SCR inlet NOx concentration includes:

[0025] Determine the initial characteristic variables affecting the SCR inlet NOx concentration. The initial characteristic variables include: the primary air-to-coal ratio of the unit, the oxygen content, the unit load, the total air volume, the total coal volume, the total air-to-coal ratio, the SOFA damper opening, the secondary damper opening, and the primary air temperature;

[0026] Define the initial characteristic variables as X = [x1(t), x2(t), …, x n (t)], and the corresponding output variable is the SCR inlet NOx concentration data actually measured at the corresponding moment, defined as c = c(t), where n represents the number of initial characteristic variables, and t represents the time of value taking;

[0027] Perform phase space reconstruction on the initial characteristic variables by embedding the time delay τ = (τ1, τ2, … τ d ) to obtain the vector [x i (t - τ1), x i (t - τ2), …, x i (t - τ d )] after embedding the delay;

[0028] Calculate the values of the mutual information MI(x i (t - τ1), c(t)), … MI(x i (t - τ d ), c(t)) respectively, and when the obtained value of the mutual information is the largest, the corresponding time delay τ h ∈[τ1, τ2, …, τ d is the pure delay between the initial characteristic variables and the SCR inlet NOx concentration.

[0029] Optionally, the process of adding pure delay to the initial feature variables to obtain intermediate variables and screening out feature variables from the intermediate variables using the maximum correlation minimum redundancy method includes:

[0030] The intermediate variable obtained by adding pure delay is defined as XD = [x1(t - τ h1 ), x1(t - τ h2 ), …, x n (t - τ hn )];

[0031] Screen out feature variables from the intermediate variables using the maximum correlation minimum redundancy method. The feature variable set is denoted as f = [x p1 (t - τ p1 ), x p2 (t - τ p2 ), …, x pm (t - τ pm )], satisfying

[0032] Optionally, the process of delaying the feature variables using the delay time and pure delay to obtain the input variables includes:

[0033] Take each variable in the set f of feature variables that affect the SCR inlet NOx concentration, which is screened out, as the initial input variable x pi (t);

[0034] Perform delay processing on the initial input variable x pi (t) to obtain the input variable x pi (t - k i ), where τ pi is the pure delay, t d is the delay time, and T is the adopted period.

[0035] Optionally, the training process of the NARX neural network includes:

[0036] Input the input variable x pi (t - k i ) into the NARX neural network to obtain the reconstructed SCR inlet NOx concentration c m (t); Delay c m (t) by 1 beat to obtain the data c m (t - 1), and c m (t - 1) is used as the feedback input signal of the NARX neural network;

[0037] Calculate the difference between c m (t) and c(t + t d )d ) is the measured NOx concentration at the SCR inlet that is t d seconds ahead of the actual measured data c(t);

[0038] When the calculated difference meets the expectation, the current NARX neural network is used as the pre-trained NARX neural network. When the calculated difference does not meet the expectation, the training parameters in the current NARX neural network are adjusted and training continues.

[0039] Optionally, the functional relationship between the measured value of the total air volume and the flue gas flow is obtained by least squares fitting, and the specific form of the function is f(x) = ax + b, where a = 0.825 and b = 30.75.

[0040] Optionally, the calculation formula for the ammonia injection amount required for the feedforward control of the flue gas denitration system is obtained by using the reconstructed NOx concentration at the SCR inlet and the flue gas flow:

[0041] F fog × (C NOx_in - C NOx_out ) / 46 × η × 17 / 10 6 = F NH3

[0042] where, F fog is the flue gas flow, C NOx_in is the reconstructed NOx concentration at the SCR inlet, C NOx_out is the set NOx concentration at the SCR outlet, η is the ammonia-nitrogen molar ratio, 46 and 17 are the molar masses of NO2 and NH3 respectively, and F NH3 is the ammonia injection amount. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The drawings described herein are used to provide a further understanding of the present invention, form a part of this application, and do not limit the present invention. In the drawings:

[0044] Figure 1 is a schematic flowchart of a feedforward control method for a flue gas denitration system based on signal reconstruction according to the present invention;

[0045] Figure 2 is a schematic diagram of the feedforward control principle of the ammonia injection amount in a feedforward control method for a flue gas denitration system based on signal reconstruction according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the embodiments and the drawings. Herein, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but do not limit the present invention.

[0047] See Figure 1 , a feedforward control method for a flue gas denitration system based on signal reconstruction according to the present invention includes:

[0048] S100, calculating the delay time of the SCR inlet NOx concentration measurement signal.

[0049] During implementation, when sampling and measuring the NOx concentration in the flue gas of a coal-fired unit, due to the long sampling pipeline, a measurement time lag problem occurs, resulting in a time lag in the feedforward control signal of the ammonia injection amount, seriously affecting the control effect. In order to ensure the real-time nature of the reconstructed NOx concentration, it is necessary to determine the delay time of the SCR inlet NOx concentration measurement signal, specifically including:

[0050] Obtaining the oxygen content data in the flue gas within a predetermined time period, which is directly measured by zirconia, and there are obvious changes in the oxygen content data within the predetermined time period;

[0051] Obtaining the SCR inlet NOx concentration data within a predetermined time period;

[0052] Embedding different pure delay times into the obtained oxygen content data to obtain the reconstructed oxygen content data, and the pure delay time can be selected as 30 seconds, 40 seconds... 120 seconds;

[0053] Calculating the person correlation coefficient between the reconstructed oxygen content data and the obtained SCR inlet NOx concentration data respectively, and the calculation formula is as follows:

[0054]

[0055] Among them, X represents the time series of the reconstructed oxygen content data, that is, O2(t - t d ), t d represents the embedded pure delay time; Y represents the time series of the SCR inlet NOx concentration data, that is, NO x (t); cov(X,Y) represents the covariance of the sequences X and Y; δ X represents the standard deviation of the sequence X, and δ Y represents the standard deviation of the sequence Y;

[0056] When the correlation coefficient ρ(X,Y) is the largest, the pure delay time t d embedded in the reconstructed oxygen content data O2(t - t d ) is used as the delay time of the SCR inlet NOx concentration measurement signal.

[0057] S110, determining the initial characteristic variables affecting the SCR inlet NOx concentration, and applying the mutual information method to determine the pure delay between the initial characteristic variables and the SCR inlet NOx concentration.

[0058] In implementation, through mechanism analysis, the initial characteristic variables affecting the SCR inlet NOx concentration are determined, including: the primary air-coal ratio of the unit, oxygen content, unit load, total air volume, total coal volume, total air-coal ratio, SOFA damper opening, secondary damper opening, and primary air temperature;

[0059] Define the initial characteristic variables as X = [x1(t), x2(t), …, x n (t)], and the corresponding output variable is the SCR inlet NOx concentration data actually measured at the corresponding time, defined as c = c(t), where n represents the number of initial characteristic variables, and t represents the time of value taking;

[0060] The time delay between each initial characteristic variable and the output variable is different. Embed the time delay τ = (τ1, τ2, … τ d ) for phase space reconstruction (the value of τ can be set according to experience), and obtain the vector after embedding the delay [x i (t - τ1), x i (t - τ2), …, x i (t - τ d )];

[0061] Calculate the mutual information MI(x i (t - τ1), c(t)), … MI(x i (t - τ d ), c(t)) respectively, and when the mutual information value obtained is the largest, the corresponding time delay τ h ∈ [τ1, τ2, …, τ d , then it is the pure time delay between the initial characteristic variable and the SCR inlet NOx concentration.

[0062] S120, add the pure time delay to the initial characteristic variable to obtain an intermediate variable, and use the maximum correlation minimum redundancy method to screen out the characteristic variables from the intermediate variables.

[0063] In implementation, the intermediate variable obtained by adding the pure time delay is defined as XD = [x1(t - τ h1 ), x1(t - τ h2 ), …, x n (t - τ hn )];

[0064] Use the maximum correlation minimum redundancy method to screen out the characteristic variables from the intermediate variables. The characteristic variable set is denoted as f = [x p1 (t - τ p1 ), x p2 (t - τ p2 ), …, x pm (t - τ pm )], satisfying

[0065] S130. Use the delay time and the pure time delay to perform a delay process on the characteristic variable to obtain the input variable.

[0066] During implementation, each variable in the set f of characteristic variables that affect the SCR inlet NOx concentration is used as the initial input variable x pi (t);

[0067] Perform a delay process on the initial input variable x pi (t) to obtain the input variable x pi (t - k i ), where τ pi is the pure time delay, t d is the delay time, and T is the adopted period.

[0068] S140. Feed the input variable into the pre-trained NARX neural network to obtain the reconstructed SCR inlet NOx concentration.

[0069] During implementation, feed the input variable x pi (t - k i ) into the NARX neural network to obtain the reconstructed SCR inlet NOx concentration c m (t); Feed the data c m (t) delayed by 1 beat to obtain c m (t - 1), and c m (t - 1) is used as the feedback input signal of the NARX neural network;

[0070] During the training of the NARX neural network, take c m (t) = c(t + t d ), so as to realize that the reconstructed SCR inlet NOx concentration c m (t) is ahead of the actual measurement signal c(t) by t d seconds, achieving the purpose of compensating for the pure time delay of the measurement signal;

[0071] During the specific training process, after feeding the input variable x pi (t - k i ) into the NARX neural network to obtain the reconstructed SCR inlet NOx concentration c m (t), the difference between c m (t) and c(t + t d ) can be calculated. c(t + t d ) is the measured SCR inlet NOx concentration that is ahead of the actual measurement data c(t) by t d seconds;

[0072] When the calculated difference meets the expectation, it indicates that the current NARX neural network meets the requirements, and the current NARX neural network is used as the pre-trained NARX neural network. When the calculated difference does not meet the expectation, the training parameters in the current NARX neural network are adjusted and training continues.

[0073] S150. Obtain the measured value of the total air volume, and use the functional relationship between the measured value of the total air volume and the flue gas flow rate to obtain the reconstructed flue gas flow rate.

[0074] In implementation, the functional relationship between the measured value of the total air volume and the flue gas flow rate is obtained by fitting using the least squares fitting method. The specific form of the function is f(x) = ax + b, where a = 0.825 and b = 30.75.

[0075] S160. According to the NOx removal principle, use the reconstructed NOx concentration at the SCR inlet and the flue gas flow rate to obtain the ammonia injection amount required for the feedforward control of the flue gas denitration system.

[0076] When the denitration system is in a steady-state condition and the ammonia slip can be ignored, it can be approximately considered that all the ammonia injection amount participates in the NOx removal reaction. Generally, it is considered that the NOx in the flue gas is mainly composed of NO and NO2, and the molar ratio of NO and NO2 in the flue gas can be approximately considered as 95:5. According to the NOx removal principle of the denitration system, the following chemical reactions occur:

[0077] 4NH3 + 4NO + O2 = 4N2 + 6H2O

[0078] 4NH3 + 2NO2 + O2 = 3N2 + 6H2O

[0079] Combined with the measuring point situation of the actual denitration system, the measuring points of the NOx concentration are arranged at the inlet and outlet of the denitration reactor. The measurement result of the NOx concentration is the converted NO2 mass concentration, that is, (the number of moles of NO per unit volume of flue gas + the number of moles of NO2 per unit volume of flue gas) × the molar mass of NO2 × the O2 conversion value. Therefore, when the SCR denitration system is in a steady state, the following formula holds:

[0080] F fog ×(C NOx_in -C NOx_out ) / 46×η×17 / 10 6 =F NH3

[0081] Among them, F fog is the flue gas flow rate, C NOx_in is the reconstructed NOx concentration at the SCR inlet, C NOx_out is the set NOx concentration at the SCR outlet, η is the ammonia-nitrogen molar ratio, 46 and 17 are the molar masses of NO2 and NH3 respectively, FNH3 is the ammonia injection amount;

[0082] The principle of feedforward control is as Figure 2 shown and is specifically described as follows: The reconstructed SCR inlet NOx concentration is used to replace the inlet NOx measurement signal in the conventional feedforward control logic. The difference between it and the set value of the SCR outlet NOx concentration is the NOx concentration value to be removed, with the unit of mg / m3. Since the flue gas flow signal measurement is inaccurate, the present invention uses the product of the reconstructed flue gas flow signal and the removed NOx concentration value as the NOx flow value to be removed, with the unit of g / h. Considering that the measured value unit of the ammonia injection amount is Kg / h, the NOx flow value to be removed is multiplied by the correction coefficient 17 / (46×1000) to obtain the corresponding ammonia injection amount feedforward control instruction, thereby ensuring that the ammonia-nitrogen molar ratio is 1.05:1.

[0083] Example:

[0084] Calculate the delay time of the SCR inlet NOx concentration measurement signal:

[0085] In the first step, extract the oxygen content data during the unit load reduction process. The time period length is 10 minutes, the minimum value of the oxygen content is 2.1, and the maximum value is 3.5. In the second step, extract the SCR inlet NOx concentration data during the same time period. The minimum value of the NOx concentration is 210 mg / m3, and the maximum value is 400 mg / m3. In the third step, according to experience, take the delay time t d as 30 seconds, 40 seconds,..., 120 seconds respectively, and calculate the person correlation coefficient between the reconstructed oxygen content signal O2(t - t d ) and the SCR inlet NOx concentration signal NO x (t). When the correlation coefficient is the largest, the corresponding delay time t d ≈90 seconds.

[0086] Determine the initial characteristic variables affecting the SCR inlet NOx concentration, and use the mutual information method to determine the pure delay between the initial characteristic variables and the SCR inlet NOx concentration:

[0087] Select the unit primary air-coal ratio, oxygen content, unit load, total air volume, total coal volume, total air-coal ratio, SOFA damper opening, secondary damper opening, and primary air temperature as the initial input variables. Considering that the SOFA damper opening and the secondary damper opening are functions of the unit load, the unit load is used instead in the mutual information analysis. Therefore, there are a total of 7 initial characteristic variables.

[0088] Define the initial characteristic variables as \(X = [x_1(t), x_2(t), \ldots, x_7(t)]\), and the corresponding output variable is the SCR inlet NOx concentration data actually measured at the corresponding moment, defined as \(c = c(t)\); the delay between each initial characteristic variable and the output variable is different. For the characteristic variable \(x\) i (t), perform phase space reconstruction, and embed the time delay \(\tau = (\tau_1, \tau_2, \ldots, \tau\) d ) Considering the data sampling period \(T = 10s\), according to experience, the minimum and maximum values of \(\tau\) are 60 seconds and 200 seconds. Therefore, take \(\tau = (60s, 70s, \ldots, 200s)\) to obtain the vector of the initial characteristic variable \(x\) i (t) after embedding the time delay as \([x\) i (t - \tau_1), x\) i (t - \tau_2), \ldots, x\) i (t - \tau\) d )]; Calculate the values of mutual information \(MI(x\) i (t - \tau_1), c(t)), \ldots, MI(x\) i (t - \tau\) d ), c(t)). The delay time \(\tau\) h \(\in [\tau_1, \tau_2, \ldots, \tau\) d corresponding to the maximum mutual information is the pure delay between the input variable \(x\) i (t) and the SCR inlet NOx concentration. The calculation results are shown in Table 1:

[0089]

[0090] Table 1 Calculation results of input variable delay

[0091] Thus, the intermediate variable after adding the pure delay is obtained, defined as:

[0092] \(X_D = [x_1(t - \tau\) h1 ), x_1(t - \tau\) h2 ), \ldots, x\) n (t - \tau\) hn )]

[0093] Use the maximum correlation and minimum redundancy method to screen out the characteristic variables that affect the SCR inlet NOx concentration from the variable set \(X_D\). When the threshold is set to 0.1, the obtained characteristic variable set is:

[0094] \(X_P = [x\) p1 (t - 90s), x\) p2 (t - 150s), x\) p3 (t - 140s), x\) p4 (t - 150s)]

[0095] Among them, \(x\)p1 , x p2 , x p3 , x p4 respectively represent the oxygen content, the primary air-to-coal ratio, the unit load, and the total coal quantity.

[0096] Reconstruction of the SCR inlet NOx concentration:

[0097] The signal x p1 (t), x p2 (t), x p3 (t), x p4 (t) is used as the input signal of the delay link, and the output signal of the delay link is x p1 (t - 90s + t d ), x p2 (t - 150s + t d ), x p3 (t - 140s + t d ), x p4 (t - 150s + t d ), t d It has been determined in (1), that is, t d = 90s; the output signal of the delay link is used as the input signal of the NARX neural network. At the same time, the output signal c m (t) of the NARX neural network is lagged by one sampling period T = 10s, and the signal c m (t - 10s) is fed back to the input end of the NARX neural network model. When training the NARX neural network model, the value of c m (t) corresponds to c(t + t d ), c(t) is the measured data of the SCR inlet NOx concentration, and t d = 90s is the measurement lag time of the SCR inlet NOx concentration. The NARX neural network model is trained and tested based on historical operation data.

[0098] Reconstruction of the flue gas flow rate:

[0099] First, based on the historical data of unit operation, the steady-state operating point data is retrieved using the steady-state search method. The data format is [total air volume, SCR inlet NOx concentration, SCR outlet NOx concentration, ammonia injection volume].

[0100] Then, the flue gas flow rate is calculated according to the following formula.

[0101] F fog ×(C NOx_in - C NOx_out ) / 46 × η × 17 / 10 6 = F NH3

[0102] Where: F fog is the flue gas flow rate, unit: m3 / h; C NOx_in is the NOx concentration at the SCR inlet, unit: mg / m3; C NOx_out is the NOx concentration at the SCR outlet, unit: mg / m3; η is the ammonia-nitrogen molar ratio, according to formulas (1) and (2), η generally takes 1.05; F NH3 is the ammonia injection rate, unit Kg / h; 46 and 17 are the molar masses of NO2 and NH3 respectively.

[0103] Furthermore, a steady-state data set is obtained, in the format of [total air volume, flue gas flow rate]. Specifically, the steady-state data set obtained from historical data, in the format of [total air volume, flue gas flow rate], through the method of least squares fitting, fits the functional relationship between the flue gas flow rate signal and the total air volume, realizing the reconstruction of the flue gas flow rate signal. The specific form of the function is f(x) = ax + b. Where a = 0.825 and b = 30.75.

[0104] Formation of the feedforward control signal:

[0105] Measure the total air volume of the data, obtain the flue gas flow rate through the function f(x) = ax + b, subtract the reconstructed NOx concentration at the SCR inlet from the set value of the NOx concentration at the SCR outlet, multiply the result of the subtraction by the flue gas flow rate, and then multiply the result of the multiplication by the correction factor 17 / (46*1000), then the feedforward control command for the ammonia injection rate is obtained.

[0106] The above are only the preferred embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention are all included in the protection scope of the present invention.

Claims

1. A feedforward control method for a flue gas denitration system based on signal reconstruction, characterized in that The method includes: Calculating the delay time of the SCR inlet NOx concentration measurement signal; Determining the initial characteristic variables affecting the SCR inlet NOx concentration, and applying the mutual information method to determine the pure delay between the initial characteristic variables and the SCR inlet NOx concentration; Adding the pure delay to the initial characteristic variables to obtain intermediate variables, and screening out the characteristic variables from the intermediate variables by using the maximum correlation minimum redundancy method; Performing delay processing on the characteristic variables by using the delay time and the pure delay to obtain input variables; Feeding the input variables into the pre-trained NARX neural network to obtain the reconstructed SCR inlet NOx concentration; Obtaining the measured value of the total air volume, and using the functional relationship between the measured value of the total air volume and the flue gas flow rate to obtain the reconstructed flue gas flow rate; According to the NOx removal principle, using the reconstructed SCR inlet NOx concentration and the flue gas flow rate to obtain the ammonia injection amount required for the feedforward control of the flue gas denitration system; The process of calculating the delay time of the SCR inlet NOx concentration measurement signal includes: Obtaining the oxygen content data in the flue gas within a predetermined time period; Obtaining the SCR inlet NOx concentration data within a predetermined time period; Embedding different pure delay times into the obtained oxygen content data to obtain the reconstructed oxygen content data; Calculating the person correlation coefficient between the reconstructed oxygen content data and the obtained SCR inlet NOx concentration data respectively, and the calculation formula is as follows: Among them, X represents the time series of the oxygen content data after reconstruction, i.e., O2(t - t d ), where t d represents the embedded pure delay time; Y represents the time series of the SCR inlet NOx concentration data, i.e., NO x (t); cov(X,Y) represents the covariance of the sequences X and Y; δ X represents the standard deviation of the sequence X, and δ Y represents the standard deviation of the sequence Y; When the correlation coefficient ρ(X,Y) is maximum, the pure delay time t embedded in the reconstructed oxygen content data O2(t - t d ) is used as the delay time of the SCR inlet NOx concentration measurement signal. d ​ 2. The feedforward control method for the flue gas denitration system based on signal reconstruction according to claim 1, characterized in that, The process of determining the initial characteristic variables affecting the SCR inlet NOx concentration and applying the mutual information method to determine the pure delay between the initial characteristic variables and the SCR inlet NOx concentration includes: Determining the initial characteristic variables affecting the SCR inlet NOx concentration, and the initial characteristic variables include: unit primary air-coal ratio, oxygen content, unit load, total air volume, total coal volume, total air-coal ratio, SOFA damper opening, secondary damper opening, primary air temperature; Define the initial feature variables as \(X = [x_1(t), x_2(t), \cdots, x n (t)]\), and the corresponding output variable is the SCR inlet NOx concentration data actually measured at the corresponding moment, defined as \(c = c(t)\), where \(n\) represents the number of initial feature variables, and \(t\) represents the time of value taking; Perform phase space reconstruction on the initial characteristic variable embedding time delay τ = (τ1, τ2, ··· τ d ) to obtain the vector [x i (t - τ1), x i (t - τ2), ···, x i (t - τ d )] after embedding delay; Calculate the mutual information MI(x i (t - τ1), c(t)), … MI(x i (t - τ d ), c(t)) values respectively, and when the mutual information value obtained is the largest, the corresponding time delay τ h ∈[τ1, τ2, …, τ d , then it is the pure time delay between the initial characteristic variable and the NOx concentration at the SCR inlet.

3. The feedforward control method for the flue gas denitration system based on signal reconstruction according to claim 2, wherein, The process of adding the pure delay to the initial characteristic variables to obtain intermediate variables and screening out the characteristic variables from the intermediate variables by using the maximum correlation minimum redundancy method includes: The intermediate variable obtained by adding pure delay is defined as XD = [x1(t - τ h1 ), x1(t - τ h2 ), …, x n (t - τ hn )]; Using the maximum correlation and minimum redundancy method to screen out the feature variables from the intermediate variables, and the set of feature variables is denoted as f = [x p1 (t - τ p1 ), x p2 (t - τ p2 ), …, x pm (t - τ pm )], which satisfies 4. The feedforward control method for a flue gas denitration system based on signal reconstruction according to claim 3, wherein, The process of performing delay processing on the characteristic variables by using the delay time and the pure delay to obtain input variables includes: Each variable in the set f of characteristic variables that affect the NOx concentration at the SCR inlet, which has been screened out, is used as the initial input variable x pi (t); Perform a delay process on the initial input variable x pi (t) to obtain the input variable x pi (t - k i ), where τ pi is the pure delay, t d is the delay time, and T is the adopted period.

5. The feedforward control method for the flue gas denitration system based on signal reconstruction according to claim 4, wherein The training process of the NARX neural network includes: Input variable x pi (t - k i ) into the NARX neural network to obtain the reconstructed NOx concentration c m (t) at the SCR inlet; delay c m (t) by one beat to obtain the data c m (t - 1), and use c m (t - 1) as the feedback input signal of the NARX neural network; Calculate c m (t) and the difference between c(t + t d ), where c(t + t d ) is the measured SCR inlet NOx concentration that is t d seconds ahead of the actual measured data c(t); When the calculated difference meets the expectation, using the current NARX neural network as the pre-trained NARX neural network; when the calculated difference does not meet the expectation, adjusting the training parameters in the current NARX neural network and continuing the training.

6. The feedforward control method for the flue gas denitration system based on signal reconstruction according to claim 1, wherein The functional relationship between the measured value of the total air volume and the flue gas flow rate is obtained by using the least squares fitting method, and the specific form of the function is f(x)=ax + b, where a = 0.825 and b = 30.

75.

7. The feedforward control method for a flue gas denitration system based on signal reconstruction according to claim 6, characterized in that The calculation formula for obtaining the ammonia injection amount required for the feedforward control of the flue gas denitration system by using the reconstructed SCR inlet NOx concentration and the flue gas flow rate is: F fog ×(C NOx_in -C NOx_out ) / 46×η×17 / 10 6 = F NH3 Among them, F fog is the flue gas flow rate, C NOx_in is the reconstructed NOx concentration at the SCR inlet, C NOx_out is the set NOx concentration at the SCR outlet, η is the ammonia-nitrogen molar ratio, 46 and 17 are the molar masses of NO2 and NH3 respectively, F NH3 is the ammonia injection amount.

Images