A control method and device for denitrification of flue gas at the tail of a cement kiln

Abstract

The present invention relates to a method for controlling denitrification of flue gas at the tail of a cement kiln, including: calculating the initial addition amount C of a reducing agent 还设 ; after both the SNCR and SCR systems are put into operation, a feedback control method is adopted to control the addition amount of the reducing agent in the SCR system, and basic data within a certain operation time is obtained. The data is randomly divided into two groups, one of which is used to train a model, and the other is used for verification; if the verification fails, the data range is expanded to obtain accurate data and continue training until a successful model is trained; the successfully trained model is used to control the operation of the SCR system, and data is continuously recorded. After a certain period of time, a new model is trained with the newly obtained data, and the newly trained model is used to control the operation of the SCR system. The present invention solves the problem that the addition amount of the reducing agent in the existing control technology does not match the actual demand, obtains the function of each relevant parameter affecting the addition amount of the reducing agent, and makes the operation of the denitrification system closer to the design value.

Description

Technical Field

[0001] The present invention relates to the technical field of waste gas treatment in the cement industry, and particularly relates to a method and device for controlling denitrification of flue gas at the tail of a cement kiln. Background Art

[0002] The dust concentration in the flue gas at the tail of a cement kiln is high. In order to reduce the abrasion, blockage and poisoning of the catalyst by dust, an electrostatic precipitator needs to be added after SNCR (Selective Non-Catalytic Reduction) and before entering SCR (Selective Catalytic Reduction). The relevant parameters at the SCR inlet are measured before the electrostatic precipitator, while the emission concentration of NOx can only be measured recently at the outlet of the SCR, and in some cases, it is even placed further downstream in the flue gas, such as in the chimney. In this way, the time for the flue gas to flow from the inlet measuring point to the outlet measuring point is dozens of seconds.

[0003] In order to ensure the economic and efficient operation of SCR, it is necessary to control the amount of reducing agent added to the SCR system well. If too much reducing agent is added, ammonia escape and waste of the reducing agent will occur; if too little reducing agent is added, the NOx emission concentration will exceed the standard.

[0004] The main factors affecting the injection amount of the reducing agent are as follows: flue gas volume (the larger this value, the more reducing agent is consumed), flue gas temperature (within the range of 300 - 380 °C, the higher the temperature, the less reducing agent is consumed), inlet NOx concentration (the higher this value, the more reducing agent is consumed), dust concentration (the content of alkali metals and alkaline earth metals in the dust in the flue gas at the tail of the cement industry kiln is much higher than that in other industries, which has a negative impact on the amount of reducing agent used), and total catalyst operation time (the longer the operation time, the lower the denitrification efficiency). Due to various factors, the flue gas volume, flue gas temperature, inlet NOx concentration, and dust concentration fluctuate within a certain range, and the operation time of the catalyst increases linearly within its service life. Only by tracking the above parameters well can SCR operate efficiently and economically.

[0005] For the control of the injection amount of the reducing agent for SCR, generally, the amount of reducing agent added at the inlet of the SCR reactor is adjusted according to the content of NOx at the outlet of the SCR reactor. This control method is called feedback control. The disadvantage of feedback control is its lag. Using data from dozens of seconds ago to control the current injection amount of the reducing agent does not conform to the actual working conditions. The amount of reducing agent added may be more or less than the actual required amount, resulting in the situation of excessive emission of NOx and the reducing agent or waste of the reducing agent. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a denitration control method and device for the flue gas at the tail of a cement kiln in view of the problem in the above-mentioned prior art that the amount of reducing agent added does not match the actual demand of the flue gas. The device controls the amount of reducing agent added according to main real-time parameters such as flue gas flow rate, inlet NOx concentration, flue gas temperature, dust concentration, and operation time of the catalyst, thus solving the problem that the amount of reducing agent added does not match the actual demand of the flue gas.

[0007] The technical solution adopted by the present invention to solve the above-mentioned technical problem is as follows:

[0008] A denitration control method for the flue gas at the tail of a cement kiln, which is applied to the SCR denitration system at the tail of a cement kiln, includes the following steps:

[0009] S1. Calculate the denitration efficiency η of the SCR system according to the type of preheater at the tail of the cement kiln and the designed concentration C NOx入设 of NOx at the inlet of the SCR and the target emission concentration C NOx出设 of NOx at the outlet of the SCR; then determine the amount of reducing agent added C NOx入设 according to the designed concentration C 还设 of NOx at the inlet of the SCR and η, and C 还设 is the initial amount of reducing agent added;

[0010] After both the SNCR and SCR systems are put into operation, adopt the method of feedback control to control the amount of reducing agent added to the SCR system. Obtain the basic data within a certain operation time through the flue gas inlet instrument and the flue gas outlet instrument, and randomly divide the data into 2 groups, one of which is used to train the model and the other is used for verification; if the verification fails, expand the data range to obtain accurate data and continue training until a successful model is trained;

[0011] S3. Control the operation of the SCR system with the successfully trained model, and continue to record data. After a certain period of time, train a new model with the newly obtained data and control the operation of the SCR system with the newly trained model.

[0012] In the above method, the step S2 specifically includes:

[0013] S2.1. Adopt the method of feedback control to control the amount of reducing agent added to the SCR system. This moment is recorded as t0, and record the data of T gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C 还 and the C NOx出 at the outlet of the SCR at (t0 + n) seconds. Among them, T gas is the flue gas temperature at the inlet of the SCR, Vgas is the flue gas flow rate at the SCR inlet, C NH3 is the NH3 concentration at the SCR inlet, C NOx入 is the NOx concentration at the SCR inlet, C 尘 is the dust concentration at the SCR inlet, t catalyst is the total operating time of the SCR catalyst, P catalyst is the system resistance, C 还 is the amount of reductant added; n is the time for the flue gas to flow from the inlet measurement point to the outlet measurement point of the SCR system, n = L / V gas , L is the equivalent length from the inlet measurement point to the outlet measurement point of the SCR; the above data form an array A0 = (C 还 , T gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 , dC NOx出 ), where dC NOx出 is the absolute value of the deviation between the actual NOx content at the outlet C NOx出 and the real-time set value C NOx实设 ; data is recorded every time interval Δt to form an array Ai at that moment, i = 0, 1, 2, 3...;

[0014] S2.2. After running for a certain period of time, sort the array Ai in ascending order according to dC NOx出 , take the first N arrays, and randomly divide them into 2 groups. The control system transmits these 2 groups of data to the computing center. One group uses machine learning methods or deep learning methods to train these arrays to find the relationship function f(T 还 , V gas , C gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ) between the amount of reductant added C gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ), and the other group is used for verification. When f(T gas , V gas , C NH3 , C NOx入 , C 尘 , tcatalyst , P catalyst , C NOx出 ) When the deviation from C 还 is less than the set value, it is considered that the model training is successful, and this model is denoted as f0;

[0015] S2.3. If the trained model cannot meet the error of the set value, still use the feed-back control method to control the operation of the SCR system, continue to record data, expand the data screening range according to step S2.2 to obtain more accurate data, and continue to train the model until the model training is successful, and this model is denoted as f0.

[0016] In the above method, the step S3 specifically includes:

[0017] S3.1. Use the successfully trained model f0 to control the operation of the SCR system. This moment is denoted as T0, and record the T gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C 还 and the data of C at the SCR outlet at (T0 + n) seconds NOx出 . The above data forms an array B0 = (C 还 , T gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 , dC NOx出 ); Record the data every time interval Δt to form an array Bi at this moment, i = 0, 1, 2, 3...;

[0018] S3.2. After running for a certain period of time, sort the array Bi in ascending order according to dC NOx出 , take out the first N arrays, and randomly divide them into 2 groups. The control system transmits these 2 groups of data to the computing center. One of the groups uses machine learning methods or deep learning methods to train these arrays to find the relationship function f(T 还 between the reducing agent addition amount C gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ) and the parameters; gas , V gas , C NH3 , CNOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ), and another set is used for verification. When the deviation between f(T gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ) and C 还 is less than the set value, it is considered that the model training is successful, and this model is denoted as f j+1 , j = 0, 1, 2, 3...;

[0019] S3.3. Use the model f j+1 to control the operation of the SCR system, and loop steps S3.1 and S3.2.

[0020] In the above method, in step S1, the denitration efficiency η = (C NOx入设 - C NOx出设 ) / C NOx入设 * 100%;

[0021] The reducing agent addition amount C 还设 = α * C NOx入设 * η, where α is an empirical coefficient.

[0022] In the above method, the system resistance P catalyst = P 出 - P 入 , where P 入 is the SCR inlet pressure, and P 出 is the SCR outlet pressure.

[0023] Correspondingly, the present invention also proposes a denitration control device for the flue gas of the cement kiln tail. The denitration control device is applied to the SCR denitration system of the cement kiln tail, and an electrostatic precipitator is provided before the SCR denitration system; the denitration control device includes a flue gas inlet instrument, a flue gas outlet instrument, a DCS control system, an execution system, and a calculation center; the flue gas inlet instrument is arranged at the inlet of the electrostatic precipitator and is used to measure the SCR inlet flue gas temperature T gas , the SCR inlet flue gas flow rate V gas , the SCR inlet NOx concentration C NOx入 , the SCR inlet NH3 concentration C NH3 , the SCR inlet dust concentration C 尘 , the total operation time t of the SCR catalyst catalyst , the SCR inlet pressure P 入 , and the reducing agent addition amount C 还; The flue gas outlet instrument is installed at the outlet of the SCR denitration system and is used to measure the actual concentration C of NOx at the SCR outlet NOx出 and the outlet pressure P 出 ; The execution system is installed at the inlet of the electrostatic precipitator and is used to spray the reducing agent; the flue gas inlet instrument and the flue gas outlet instrument are respectively connected to the DCS control system, and the DCS control system is respectively connected to the execution system and the calculation center; the data measured by the flue gas inlet instrument and the flue gas outlet instrument are stored in the DCS control system, the DCS control system records the above data, calculates the system resistance, transmits the relevant data to the calculation center according to the program, the calculation center receives the data transmitted by the DCS control system, and trains the model based on these data, and the DCS control system downloads the successfully trained model from the calculation center to control the execution system.

[0024] In the above solution, the execution system is an ammonia injection device.

[0025] The beneficial effects of the present invention are as follows:

[0026] 1. The device of the present invention installs a flue gas inlet instrument at the inlet of the electrostatic precipitator and a flue gas outlet instrument at the outlet of the SCR denitration system, and controls the addition amount of the reducing agent according to the actual operation parameters such as the flue gas volume, flue gas temperature, dust concentration, catalyst operation time, inlet NOx concentration, and outlet NOx concentration, solving the problem that the addition amount of the reducing agent does not match the actual demand of the flue gas.

[0027] 2. The present invention uses the method of deep learning to obtain the function of each relevant parameter affecting the addition amount of the reducing agent, making the denitration system closer to the design value, neither wasting the reducing agent nor achieving the NOx emission concentration and ammonia slip index.

[0028] 3. The present invention uses rolling data as the model training data, real-time tracks the operation status of the SCR system, continuously updates and optimizes the control model, makes the model more in line with the working conditions, and operates more economically. Description of the Drawings

[0029] The present invention will be further described below in conjunction with the drawings and embodiments. In the drawings:

[0030] Figure 1 is a schematic structural diagram of the cement kiln tail flue gas denitration control device of the present invention.

[0031] In the figure: 1. Flue gas inlet instrument; 2. Execution system; 3. Calculation center; 4. DCS control system; 5. Flue gas outlet instrument; 6. Electrostatic precipitator; 7. SCR denitration system; 71. Denitration catalyst. Detailed Embodiments

[0032] To have a clearer understanding of the technical features, objectives, and effects of the present invention, the specific implementation manners of the present invention will now be described in detail with reference to the accompanying drawings.

[0033] The present invention provides a method for controlling denitrification of flue gas at the tail of a cement kiln, which is applied to the SCR denitrification system at the tail of the cement kiln and includes the following steps:

[0034] S1. According to the type of preheater at the tail of the cement kiln, the designed concentration C NOx入设 of NOx at the SCR inlet, and the target emission concentration C NOx出设 of NOx at the SCR outlet, calculate the denitrification efficiency η of the SCR system; then determine the reductant addition amount C NOx入设 based on the designed concentration C 还设 of NOx at the SCR inlet and η. C 还设 is the initial reductant addition amount.

[0035] Among them, the denitrification efficiency η = (C NOx入设 - C NOx出设 ) / C NOx入设 * 100%;

[0036] The reductant addition amount C 还设 = α * C NOx入设 * η, where α is an empirical coefficient.

[0037] S2. After both the SNCR and SCR systems are put into operation, adopt the method of feedback control to control the reductant addition amount of the SCR system. Obtain the basic data within a certain operation time through the flue gas inlet instrument and the flue gas outlet instrument, and randomly divide the data into 2 groups. One group is used to train the model, and the other group is used for verification; if the verification fails, expand the data range to obtain accurate data and continue training until a successful model is trained because the data is necessary and sufficient, and a successful model can definitely be trained. The specific implementation steps are as follows:

[0038] S2.1. Adopt the method of feedback control to control the reductant addition amount of the SCR system. This moment is recorded as t0, and record the data of T gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C 还 and the C NOx出 at the SCR outlet at (t0 + n) seconds. Among them, T gas is the flue gas temperature at the SCR inlet, V gas is the flue gas flow rate at the SCR inlet, C NH3 is the NH3 concentration at the SCR inlet, C NOx入 is the NOx concentration at the SCR inlet, C尘 is the dust concentration at the SCR inlet, t catalyst is the total operating time of the SCR catalyst, P catalyst is the system resistance, P catalyst = P 出 - P 入 , where P 入 is the SCR inlet pressure, P 出 is the SCR outlet pressure, C 还 is the reductant addition amount; n is the time for the flue gas to flow from the inlet measurement point to the outlet measurement point of the SCR system, n = L / V gas , L is the equivalent length from the SCR inlet measurement point to the outlet measurement point, so n varies according to V gas ; the above data forms an array A0 = (C 还 , T gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 , dC NOx出 ), dC NOx出 is the absolute value of the deviation between the actual NOx content C NOx出 at the outlet and the real-time set value C NOx实设 ; after 5s, the data is recorded once, denoted as A1, and so on. i = 0, 1, 2, 3......, every time interval Δt = 5s, the data is recorded once to form the array Ai at that moment;

[0039] S2.2. After every 48 hours of operation (which can be optimized), the arrays Ai are sorted in ascending order according to dC NOx出 , the first 20,000 (which can be optimized) arrays are taken out, and the 20,000 arrays are randomly divided into 2 groups. The control system transmits these 2 groups of data to the computing center. One group uses machine learning methods or deep learning methods to train these arrays to find the relationship function f(T 还 , V gas , C gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ) between the reductant addition amount C gas and the parameters (T gas , V NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ), and the other group is used for verification. When f(Tgas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ), when the deviation from C 还 is less than 5%, the model training is considered successful, and this model is denoted as f0;

[0040] S2.3. If the trained model cannot meet the 5% error, still use the feed-back control method to control the operation of the SCR system, continue to record data, expand the data screening range according to step S2.2 to obtain more accurate data, and continue to train the model until the model training is successful. This model is denoted as f0.

[0041] S3. Use the successfully trained model to control the operation of the SCR system, and continue to record data. After a certain period of time, retrain a model with the newly obtained data, and use the newly trained model to control the operation of the SCR system. The specific implementation steps are as follows:

[0042] S3.1. Use the successfully trained model f0 to control the operation of the SCR system. This moment is denoted as T0, and record T gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C 还 and the data of C at the SCR outlet at (T0 + n) seconds NOx出 . The above data forms an array B0 = (C 还 , T gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 , dC NOx出 ); After 5s, record the data once, denoted as B1, and so on. i = 0, 1, 2, 3..., record the data once every time interval Δt to form the array Bi at this moment;

[0043] S3.2. After running for 168 hours (which can be optimized) for a certain period of time, sort the array Bi in ascending order according to dC NOx出 , take out the first 20,000 (which can be optimized) arrays, randomly divide the 20,000 arrays into 2 groups, and the control system transfers these 2 groups of data to the computing center. One of the groups uses machine learning methods or deep learning methods to train these arrays to find the reducing agent addition amount C 还Relationship function f(T gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ), and another set is used for verification. When the deviation between f(T gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ) and C gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ) is less than 5%, it is considered that the model training is successful, and this model is denoted as f 还 , j = 0, 1, 2, 3...; j+1

[0044] S3.3. Use the model f j+1 to control the operation of the SCR system, and loop steps S3.1 and S3.2.

[0045] Figure 1 As shown, the present invention also provides a denitration control device for the flue gas at the tail of a cement kiln, which is applied to the SCR denitration system at the tail of the cement kiln. An electrostatic precipitator 6 is provided before the SCR denitration system 7. The denitration control device for the flue gas at the tail of the cement kiln of the present invention includes a flue gas inlet instrument 1, a flue gas outlet instrument 5, a DCS control system 4, an execution system 2, and a computing center 3.

[0046] The flue gas inlet instrument 1 is arranged at the inlet of the electrostatic precipitator 6 and is used to measure the SCR inlet flue gas temperature T gas , the SCR inlet flue gas flow rate V gas , the SCR inlet NOx concentration C NOx入 , the SCR inlet NH3 concentration C NH3 , the SCR inlet dust concentration C 尘 , the total operation time t of the SCR catalyst catalyst , the SCR inlet pressure P 入 , the reducing agent addition amount C 还 .

[0047] The flue gas outlet instrument 5 is arranged at the outlet of the SCR denitration system 7 and is used to measure the actual SCR outlet NOx concentration C NOx出 and the outlet pressure P出 。

[0048] The execution system is arranged at the inlet of the electrostatic precipitator 6 and is used to spray the reducing agent.

[0049] The flue gas inlet instrument 1 and the flue gas outlet instrument 5 are respectively connected to the DCS control system 4. The DCS control system 4 is respectively connected to the execution system 2 and the computing center 3. The data measured by the flue gas inlet instrument 1 and the flue gas outlet instrument 5 are stored in the DCS control system 4. The DCS control system 4 records the above data and calculates the system resistance P catalyst = P 出 - P 入 , and according to the program, transmits the relevant data to the computing center 3. The computing center 3 receives the data transmitted by the DCS control system 4 and uses these data to train the model. The DCS control system 4 downloads the successfully trained model from the computing center 3 for controlling the execution system 2.

[0050] Specifically, the execution system 2 is an ammonia injection device. Multiple layers of denitration catalysts 71 are arranged in the SCR denitration system 7 along the gas flow direction.

[0051] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the spirit and scope protected by the claims of the present invention. These all fall within the protection scope of the present invention.

Claims

1. A control method for denitrification of cement kiln tail gas, characterized in that, Applied to the SCR denitration system at the tail of a cement kiln, it includes the following steps: S1. Calculate the denitration efficiency η of the SCR system according to the type of preheater at the kiln tail of the cement kiln and the designed concentration C of NOx at the SCR inlet; then determine the reductant addition amount C NOx入设 according to the designed concentration C of NOx at the SCR inlet NOx出设 and the target emission concentration C of NOx at the SCR outlet; then determine the reductant addition amount C NOx入设 according to the designed concentration C of NOx at the SCR inlet 还设 and η; C 还设 is the initial addition amount of the reductant; After both the SNCR and SCR systems are put into operation, the addition amount of the reductant in the SCR system is controlled by the method of feedback control. Basic data within a certain operation time is obtained through the flue gas inlet instrument and the flue gas outlet instrument, and the data is randomly divided into two groups. One group is used to train the model, and the other group is used for verification. If the verification fails, the data range is expanded to obtain accurate data and continue training until a successful model is trained. Step S2 specifically includes: S2.

1. Adopt a feed-back control method to control the reductant addition amount of the SCR system. This moment is recorded as t0, and record T at moment t0 gas 、V gas 、C NH3 、C NOx入 、C 尘 、t catalyst 、P catalyst 、C 还 and the data of C at the SCR outlet at (t0 + n) seconds, where T NOx出 is the flue gas temperature at the SCR inlet, V gas is the flue gas flow rate at the SCR inlet, C gas is the NH3 concentration at the SCR inlet, C NH3 is the NOx concentration at the SCR inlet, C NOx入 is the dust concentration at the SCR inlet, t 尘 is the total operating time of the SCR catalyst, P catalyst is the system resistance, C catalyst is the reductant addition amount; n is the time for the flue gas to flow from the inlet measuring point of the SCR system to the outlet measuring point, n = L / V 还 , L is the equivalent length from the inlet measuring point to the outlet measuring point of the SCR; the above data forms an array A0 = (C gas , T 还 , V gas , C gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 , dC NOx出 ), where dC NOx出 is the absolute value of the deviation between the actual NOx content C NOx出 at the outlet and the real-time set value C NOx实设 ; Record the data every time interval Δt to form an array Ai at this moment, i = 0, 1, 2, 3...; S2.

2. After running for a certain period of time, sort the array Ai in ascending order according to dC NOx出 Take out the first N arrays and randomly divide them into two groups. The control system transmits these two groups of data to the computing center. One group uses machine learning methods or deep learning methods to train these arrays to find the relationship function f(T 还 , V gas , C gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ) between the reducing agent addition amount C gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ). The other group is used for verification. When the deviation between f(T gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ) and C 还 is less than the set value, it is considered that the model training is successful, and this model is denoted as f0; S2.

3. If the trained model cannot meet the error of the set value, still use the method of feedback control to control the operation of the SCR system, continue to record data, expand the data screening range according to step S2.2 to obtain more accurate data, and continue to train the model until the model training is successful. This model is denoted as f0; S3. Use the successfully trained model to control the operation of the SCR system and continue to record data. After a certain period of time, train a new model with the newly obtained data and use the newly trained model to control the operation of the SCR system. Step S3 specifically includes: S3.

1. Use the successfully trained model f0 to control the operation of the SCR system. This moment is denoted as T0, and record the values of T gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C 还 and the data at the SCR outlet at (T0 + n) seconds, C NOx出 . The above data forms an array B0 = (C 还 , T gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 , dC NOx出 ). Record the data once every time interval Δt to form an array Bi at this moment, where i = 0, 1, 2, 3...; S3.

2. After running for a certain period of time, sort the array Bi in ascending order according to dC NOx出 Take the first N arrays, randomly divide them into two groups, and the control system transmits these two groups of data to the computing center. One group uses machine learning methods or deep learning methods to train these arrays to find the relationship function f(T 还 , V gas , C gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ) between the amount of reducing agent added C gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ). The other group is used for verification. When the deviation between f(T gas , V gas , C NH3 , C NOx入 , C 尘 , t catalyst , P catalyst , C NOx出 ) and C 还 is less than the set value, it is considered that the model training is successful. This model is denoted as f j+1 , j = 0, 1, 2, 3...; S3.

3. Use the model f j+1 to control the operation of the SCR system, and loop through steps S3.1 and S3.

2.

2. The cement kiln tail flue gas denitration control method according to claim 1, characterized in that In step S1, the denitrification efficiency η = (C NOx入设 - C NOx出设 ) / C NOx入设 * 100%; Reductant addition amount C 还设 = α * C NOx入设 * η, where α is an empirical coefficient.

3. The cement kiln tail flue gas denitration control method according to claim 1, characterized in that, The system resistance P catalyst = P 出 - P 入 , where P 入 is the SCR inlet pressure, and P 出 is the SCR outlet pressure.

Images