SCR Denitrification Ammonia Injection Control Method for Thermal Power Plant
Through technical means such as cascade controller, ammonia flow intelligent feedforward controller and state observer, the nonlinear and large hysteresis problems of the SCR denitrification ammonia spray control system in the thermal power plant are solved, and high-precision NOx concentration control and Improved automatic operation efficiency.
Patent Information
- Application Number
- CN202411148822.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-21
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-08-21
AI Technical Summary
The SCR denitrification ammonia spray control system in thermal power plants has nonlinear and large hysteresis problems, resulting in increased ammonia consumption, equipment corrosion, and NOx emissions that do not meet the standards. The existing control solutions cannot achieve high-quality fully automated and precise control.
Adopt a cascade controller and an ammonia flow intelligent feedforward controller, combined with a Smith estimator and a state observer, using the GPC main loop and secondary loop controller, control through the state feedback and output feedback gain matrix, and introduce a competitive type The neural network model performs feedforward compensation to identify working condition parameters and NOx concentration.
It effectively solves the nonlinear and large hysteresis problems of the SCR denitrification ammonia spray control system, improves the automatic operation efficiency and ammonia spray control quality of the generator set, and realizes high-precision customized control of NOx concentration.
Smart Images

Figure CN118655866B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of industrial automation control, and in particular relates to a method for controlling SCR denitration and ammonia injection in a thermal power plant. Background Art
[0002] In actual production, if excessive ammonia is sprayed in the selective catalytic reduction (SCR) denitrification ammonia spray control system of thermal power plants, not only will the ammonia consumption increase and thus the operating cost increase, but also the increase in ammonia escape will cause a large amount of viscous ammonium bisulfate (ABS) to be produced, the cold end of the air preheater will be corroded by dust accumulation, and the economic performance of the unit equipment will be significantly reduced. If the amount of ammonia sprayed is insufficient, the outlet nitrogen oxides (NO x ) Emissions do not meet standards.
[0003] In industrial process control, there are currently NO x The emission value fluctuates greatly, the traditional proportional-integral-differential (PID) control method causes control lag, the control strategy is simple, it is impossible to achieve high-quality fully automated ammonia injection, and only considers the static characteristics of the control object and poor adaptive ability. At present, it is still impossible to achieve accurate online measurement of important parameter variables of the reaction process. Therefore, the controlled object of the SCR denitration ammonia injection control system is a nonlinear, time-varying, large lag, and large inertia process. Summary of the invention
[0004] An object of the present invention is to provide a method for controlling SCR denitration and ammonia injection in a thermal power plant, which effectively solves the nonlinearity and large hysteresis problems of the SCR denitration and ammonia injection control system in a thermal power plant.
[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a SCR denitrification ammonia injection control method for a thermal power plant, which is applied to an SCR denitrification ammonia injection control system. The SCR denitrification ammonia injection control system includes a cascade controller and an ammonia flow intelligent feedforward controller. The forward path of the inner unit negative feedback control loop in the cascade controller is composed of a sub-loop controller and an ammonia injection system, and the outer unit negative feedback control loop is composed of an inner unit negative feedback control loop plus a GPC main loop controller and a denitrification system.
[0006] The GPC main loop controller and the sub-loop controller both adopt PID control scheme, and the large hysteresis problem existing in the denitrification system is solved by Smith predictor scheme.
[0007] The SCR denitration and ammonia injection control system adopts a state observer to replace the state data of the SCR denitration and ammonia injection control system. The state observer includes a state feedback gain matrix and an output feedback gain matrix to be designed.
[0008] The output signal of the ammonia flow intelligent feedforward controller is applied between the GPC main loop controller and the sub-loop controller to identify and calculate the operating parameters of the generator set, which operating parameters include but are not limited to load, total coal volume and total air volume. The output signal of the ammonia flow intelligent feedforward controller includes the NO at the SCR inlet, SCR outlet and chimney inlet after parameter identification and adjustment. x Concentration, N2 flow and equipment instrument failure information.
[0009] Furthermore, assume that the transfer function of the secondary loop controller is , the transfer function of the ammonia injection system is , then the equivalent transfer function of the internal unit negative feedback control loop is : , It is obtained based on actual test data after a step disturbance experiment when the SCR denitrification and ammonia injection control system is running stably.
[0010] Furthermore, assuming that the transfer function of the Smith predictor is In the distributed control system workstation, a subtraction module is used to compare the output measurement signal of the master control object with the The output signal of the master object and Apply the same ammonia injection flow input disturbance signal. If the output measurement signal of the main control object is If the output signal after subtracting the output signal of is close to zero, it is determined that the transfer function of the Smith predictor fully matches the dynamics of the actual master control object.
[0011] Furthermore, the state feedback gain matrix Determined by the following algorithm: First, the state equation and state feedback control law of the SCR denitration and ammonia injection control system are listed. , , , where represents the state variable, for The first derivative of Indicates the output, Indicates input, is the state matrix, is the deviation, is the input matrix, is the output matrix.
[0012] If the SCR denitration ammonia injection control system is controllable according to, but not limited to, rank criteria and PBH rank criteria, the system closed-loop poles are configured according to the target requirements of the set enterprise production needs. , find the hope polynomial : , Indicates The closed-loop poles of the system, represents the number of closed-loop poles of the system, and then calculates the characteristic polynomial of the control system of the state observer : , where It means to calculate a determinant. is the unit matrix, is the independent variable of Laplace transform, compare and , so that the coefficients of the two are equal, thus determining the state feedback gain matrix .
[0013] Output Feedback Gain Matrix The calculation method is as follows: If the SCR denitration ammonia injection control system is observable, the system closed-loop pole is configured according to the target requirements of the set enterprise production needs. , find the hope polynomial : , and then calculate the characteristic polynomial of the control system of the state observer : ,Compare and , so that the coefficients of the two are equal, thus determining the output feedback gain matrix .
[0014] Furthermore, the transfer function of the GPC main loop controller is , using suitable engineering applications The incremental step GPC algorithm, based on the generalized minimum variance self-correcting control, absorbs the optimization strategy of multi-step prediction in predictive control and introduces the step factor Avoiding matrix inversion can solve the difficult problem of online solution in self-correcting control and the unexpected numerical ill-conditioned problem caused by matrix non-invertibility.
[0015] The controlled autoregressive moving average CARMA model of the assumed object is described by the following discrete difference equation: , where Indicates the time series Sample values, and Represent the input and output sequences of the object respectively, is zero-mean white noise, represents the independent variable of the discrete difference equation, represents an operator that shifts back one term, Indicates backward movement The operator of the term, , and are polynomials of the backshift operator of the form: , , , where , and represent the coefficients of each polynomial term, represents an operator that shifts two terms backward, Indicates backward movement The operator of the term, Indicates backward movement The operator of the term, Indicates backward movement The operator of the term, express No. The coefficient of the term, express No. The coefficient of the term, express No. The coefficient of the term.
[0016] First set , and assumes a set value or reference sequence , The following performance objective function is used for GPC optimization: , where represents the performance objective function, Because until the time The object output obtained by the input and output data The optimal prediction value is one step ahead. and are the minimum prediction time domain and the maximum prediction time domain respectively, represents the increment of input, To control the time domain, and are the output prediction deviation and control increment weighted sequences, respectively, It represents the future reference trajectory of the system output, which is the product of the set value after online softening.
[0017] make , , ,but , , set adjustable parameters ,make , where represents the input matrix, represents the output matrix, represents the initial output matrix, and denotes parameter vectors, Indicates the number of parameters. Indicates parameters, for The matrix-vector form of .
[0018] Furthermore, the feedforward compensation algorithm corresponding to the ammonia flow intelligent feedforward controller adopts a competitive neural network model. The competitive neural network model has two layers, namely an output layer and an input layer. Only one output neuron wins each calculation. The winning output neuron is marked as 1, and the other output neurons are marked as 0.
[0019] When the training of the competitive neural network model begins, the weights between the input layer and the output layer are set arbitrarily, and the min-max normalization is used for data processing. , where is the normalized data, is the minimum value of the data before normalization, is the maximum value in the data before normalization.
[0020] The root mean square error (RMSE) and mean absolute percentage error (MAPE) were used to verify the accuracy of the competitive neural network model. , , where NO x The concentration of Measured values, NO x The concentration of Prediction values, is the sample size of the study.
[0021] Furthermore, the NO in the SCR denitrification ammonia injection control system x The measurement distortion disturbance generated by the concentration analyzer during maintenance is replaced by the predicted value calculated by the competitive neural network model. x Detection values of the concentration analyzer during maintenance.
[0022] Compared with the prior art, the beneficial technical effect of the present invention is: the present invention effectively solves the nonlinearity and large hysteresis problems of the SCR denitrification ammonia injection control system of the thermal power plant by optimizing and improving the cascade controller and the ammonia flow intelligent feedforward controller, improves the automatic commissioning efficiency and ammonia injection control quality of the generator set, and achieves a higher customized control accuracy of the NOx concentration of the SCR outlet flue gas. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Figure 11 is a control structure diagram of the SCR denitration and ammonia injection control system of the present invention. In the diagram, the negative sign represents negative feedback, and the hollow circle represents the comparison point. DETAILED DESCRIPTION
[0024] A method for controlling SCR denitration and ammonia injection in a thermal power plant is applied to a SCR denitration and ammonia injection control system. The SCR denitration and ammonia injection control system includes a cascade controller, an ammonia flow intelligent feedforward controller and a NO x The forward path of the inner unit negative feedback control loop in the cascade controller is composed of the sub-loop controller and the ammonia injection system, and the outer unit negative feedback control loop is composed of the inner unit negative feedback control loop plus the GPC main loop controller and the denitrification system.
[0025] The GPC main loop controller and the sub-loop controller both adopt PID control scheme, and the large hysteresis problem in the denitration system is solved by the Smith predictor scheme. The SCR denitration ammonia injection control system adopts a state observer to replace the state data of the SCR denitration ammonia injection control system. The state observer is an estimation system with the same mathematical model, control structure diagram and initial state vector as the original physical system. The state observer includes the state feedback gain matrix and output feedback gain matrix to be designed.
[0026] The output signal of the ammonia flow intelligent feedforward controller is applied between the GPC main loop controller and the sub-loop controller to identify and calculate the load, total coal volume, total air volume and other operating parameters of the generator set. The output signal of the ammonia flow intelligent feedforward controller includes the NO x Concentration, N2 flow and equipment instrument failure, etc.
[0027] The specific method of parameter identification and tuning is as follows: (1) Assume that the transfer function of the sub-loop controller is , the transfer function of the ammonia injection system is , then the equivalent transfer function of the internal unit negative feedback control loop is : , The result is obtained from the actual test data after the step disturbance experiment is carried out when the SCR denitration and ammonia injection control system is running stably. The secondary loop controller is a proportional controller, which can set an initial value first, and then obtain the appropriate parameters through PID parameter tuning.
[0028] (2) Assume that the transfer function of the Smith predictor is , which is used to solve the nonlinear problem caused by the pure lag link in the model. In engineering applications, a subtraction module is used in the distributed control system (DCS) workstation to compare the output measurement signal of the master control object with the The output signal of the master object and Apply the same ammonia injection flow input disturbance signal. If the output measurement signal of the main control object is If the output signal after subtracting the output signal of has a small fluctuation range around zero, it can be considered that the transfer function of the Smith predictor fully matches the actual dynamics of the master control object.
[0029] (3) The actual state variables of the main control object cannot generally be directly measured and obtained. It is necessary to use a model to design a state observer to estimate the state variables and achieve state reconstruction.
[0030] State feedback gain matrix Determined by the following algorithm: First, the state equation and state feedback control law of the SCR denitration and ammonia injection control system are listed. , , , where represents the state variable, for The first derivative of Indicates the output, Indicates input, is the state matrix (also called the system matrix), is the deviation, is the input matrix (also called the control matrix), is the output matrix (also called the observation matrix).
[0031] If the SCR denitration ammonia injection control system is judged according to the relevant rank criteria, PBH rank criteria, etc. Controllable, configure the system closed loop poles according to the set enterprise production needs , find the hope polynomial : , Indicates The closed-loop poles of the system, represents the number of closed-loop poles of the system. Then calculate the characteristic polynomial of the control system of the state observer : , where It means to calculate a determinant. is the unit matrix, is the independent variable of Laplace transform, compare and , so that the coefficients of the two are equal, thus determining the state feedback gain matrix .
[0032] Output Feedback Gain Matrix The calculation method is as follows: If the SCR denitration ammonia injection control system is observable, the system closed-loop pole is configured according to the target requirements of the set enterprise production needs. , find the hope polynomial : , and then calculate the characteristic polynomial of the control system of the state observer : , where is the unit matrix, is the independent variable of Laplace transform, compare and , so that the coefficients of the two are equal, thus determining the output feedback gain matrix .
[0033] (4) The transfer function of the GPC main loop controller is: The present invention adopts a suitable engineering application The incremental step GPC (generalized predictive control) algorithm can fully absorb the optimization strategy of multi-step prediction in predictive control based on generalized minimum variance self-correcting control. At the same time, the present invention introduces the step factor Avoiding matrix inversion can solve the huge potential safety hazards in actual production, such as the difficulty of online solution caused by large amount of calculation in self-correcting control and unexpected numerical pathological problems caused by matrix irreversibility.
[0034] The controlled autoregressive moving average CARMA model of the assumed object is described by the following discrete difference equation: , where Indicates the time series Sample values, and Represent the input and output sequences of the object respectively, is zero-mean white noise, represents the independent variable of the discrete difference equation, represents an operator that shifts back one term, Indicates backward movement The operator of the term, , and are polynomials of the backshift operator of the form: , , , where , and represent the coefficients of each polynomial term, represents an operator that shifts two terms backward, Indicates backward movement The operator of the term, Indicates backward movement The operator of the term, Indicates backward movement The operator of the term, express No. The coefficient of the term, express No. The coefficient of the term, express No. The coefficient of the term.
[0035] In actual design, you can first set , and assumes a set value or reference sequence , The following performance objective function is used for GPC optimization: , where represents the performance objective function, Because until the time The object output obtained by the input and output data The optimal prediction value is one step ahead. and are the minimum prediction time domain and the maximum prediction time domain respectively, represents the increment of input, To control the time domain, and are the output prediction deviation and control increment weighted sequences, respectively, It represents the future reference trajectory of the system output, which is the product of the set value after online softening.
[0036] make , , ,but , , set adjustable parameters ,make , where represents the input matrix, represents the output matrix, represents the initial output matrix, and denotes parameter vectors, Indicates the number of parameters. Indicates parameters, for The matrix-vector form of .
[0037] (5) The feedforward compensation algorithm corresponding to the ammonia flow intelligent feedforward controller adopts a competitive neural network model. The competitive neural network model has two layers, namely the output layer and the input layer. The output layer is the core layer. Only one output neuron wins each calculation. The winning output neuron is marked as 1, and the other output neurons are marked as 0.
[0038] When the competitive neural network model training starts, the weights between the input layer and the output layer can be set arbitrarily. Because the SCR denitrification operation data are of various types and the ranges vary greatly, the present invention adopts min-max normalization and standardized data processing. , where is the normalized data, is the minimum value of the data before normalization, is the maximum value in the data before normalization.
[0039] The root mean square error (RMSE) and mean absolute percentage error (MAPE) were used to verify the accuracy of the competitive neural network model. , , where NO x The concentration of Measured values, NO x The concentration of Prediction values, is the sample size of the study.
[0040] (6) NO in SCR denitrification ammonia injection control system x The measurement distortion disturbance generated by the concentration analyzer during maintenance is replaced by the predicted value calculated by the competitive neural network model. x Detection values of the concentration analyzer during maintenance.
[0041] like Figure 1 As shown, the SCR denitration ammonia injection control system of the present invention is composed of an optimized and improved cascade controller and an ammonia flow intelligent feedforward controller. The input of the SCR denitration ammonia injection control system is NO x Concentration setting value, the output of SCR denitration ammonia injection control system is chimney inlet NO xconcentration. The forward path of the inner unit negative feedback control loop in the cascade controller is composed of the sub-loop controller and the ammonia injection system, and the outer unit negative feedback control loop is composed of the inner unit negative feedback control loop plus the GPC main loop controller and the denitrification system. Both the GPC main loop controller and the sub-loop controller are PID control schemes. The large hysteresis problem in the denitrification system is solved by the Smith predictor solution. Since some states in the SCR control system of the thermal power generation layer may be inconvenient or impossible to observe, the SCR denitrification ammonia injection control system uses a state observer to replace the state data of the actual system. The state observer includes a state feedback gain matrix to be designed and an output feedback gain matrix. The use of a state observer can make the error between the estimated value and the actual value of the state observer within a reasonable range. The output signal of the ammonia flow intelligent feedforward controller is applied between the GPC main loop controller and the sub-loop controller, mainly used to identify and calculate the load, total coal volume, total air volume and other operating parameters of the generator set, specifically including the NO at the SCR inlet, SCR outlet and chimney inlet after parameter identification and adjustment. x Concentration, N2 flow and equipment instrument failure information. Figure 1 The negative sign in the middle represents negative feedback, and the hollow circle represents the comparison point, also called the integration point, which is where two signals are compared, that is, addition or subtraction operations are performed.
[0042] The following is further described in detail with reference to specific embodiments.
[0043] Example 1: The π-type furnace system equipped with the No. 2 unit of a power plant according to the present invention is modified and applied, and the SCR denitration ammonia injection control system of the unit is adjusted according to the inlet NO x Emissions: 290mg / Nm 3 And 85% denitrification efficiency for scheme design and equipment configuration.
[0044] During the period when the unit was reduced from 1000MW to 500MW, the chimney inlet NO x The concentration setting value is always 18mg / Nm 3 , NO x The maximum deviation between the measured concentration and the set value before the modification was 31.37 mg / Nm 3 , and after the modification according to the present invention, NO x The deviation between the measured concentration value and the set value is in good agreement, with only a large deviation during the grinding stop phase, where the maximum measured value reached 26.61 mg / Nm 3 , the minimum measured value reaches 12.3mg / Nm 3 Before the transformation, the automatic operation efficiency of the unit was low, and it was basically adjusted manually; after the transformation, the automatic operation efficiency was not less than 99.35%, and only at the unit inlet NO xUnder relatively bad operating conditions such as non-routine maintenance and other conditions with extremely large concentration fluctuations, the operator needs to remove the SCR denitration ammonia injection control system of the present invention and adjust it manually. It can be seen that the ammonia injection control quality of the unit is good after adopting the SCR denitration ammonia injection control system and method of the present invention.
[0045] Calculate according to the relative deviation calculation formula: , where , , and are dimensionless values, Relative deviation, represents the number of samples, represents the sample mean, Indicates The calculation shows that the relative deviation of the concentration fluctuation at the outlet of the SCR denitration reactor is about 16.96%, which proves that the operation of the SCR denitration ammonia injection control system and method of the present invention can indirectly achieve a higher SCR outlet flue gas NO x The concentration is customized with high control accuracy and the effect is good.
[0046] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.
Claims
1. A method for controlling the injection of ammonia in SCR denitrification in a thermal power plant. It is characterized in that Applied to SCR denitration ammonia injection control system, the SCR denitration ammonia injection control system includes a cascade controller and an ammonia flow intelligent feedforward controller, the forward path of the inner unit negative feedback control loop in the cascade controller is composed of a sub-loop controller and an ammonia injection system, and the outer unit negative feedback control loop in the cascade controller is composed of an inner unit negative feedback control loop plus a GPC main loop controller and a denitration system; The GPC main loop controller and the sub-loop controller both adopt PID control scheme, and the large hysteresis problem existing in the denitration system is solved by Smith predictor scheme; The SCR denitration and ammonia injection control system adopts a state observer to replace the state data of the SCR denitration and ammonia injection control system, and the state observer includes a state feedback gain matrix and an output feedback gain matrix to be designed; The output signal of the ammonia flow intelligent feedforward controller is applied between the GPC main loop controller and the sub-loop controller to identify and calculate the operating parameters of the generator set, which include load, total coal volume and total air volume. The output signal of the ammonia flow intelligent feedforward controller includes the NO at the SCR inlet, SCR outlet and chimney inlet after parameter identification and adjustment. x Concentration, N 2 Flow and equipment instrument failure information; State feedback gain matrix Determined by the following algorithm: First, the state equation and state feedback control law of the SCR denitration and ammonia injection control system are listed. , , , where represents the state variable, for The first derivative of Indicates the output, Indicates input, is the state matrix, is the deviation, is the input matrix, is the output matrix; If the SCR denitration ammonia injection control system is controllable according to the rank criterion or the PBH rank criterion, the system closed-loop poles are configured according to the target requirements of the set enterprise production needs. , find the hope polynomial : , Indicates The closed-loop poles of the system, represents the number of closed-loop poles of the system, and then calculates the characteristic polynomial of the control system of the state observer : , where It means to calculate a determinant. is the unit matrix, is the independent variable of Laplace transform, compare and , so that the coefficients of the two are equal, thus determining the state feedback gain matrix ; Output Feedback Gain Matrix The calculation method is as follows: If the SCR denitration ammonia injection control system is observable, the system closed-loop pole is configured according to the target requirements of the set enterprise production needs. , find the hope polynomial : , and then calculate the characteristic polynomial of the control system of the state observer : ,Compare and , so that the coefficients of the two are equal, thus determining the output feedback gain matrix ; The transfer function of the GPC main loop controller is: , using suitable engineering applications The incremental step GPC algorithm, based on the generalized minimum variance self-correcting control, absorbs the optimization strategy of multi-step prediction in predictive control and introduces the step factor Avoid matrix inversion, thus solving the difficult problem of online solution in self-correcting control and the unexpected numerical ill-conditioned problem caused by non-invertible matrix; The controlled autoregressive moving average CARMA model of the assumed object is described by the following discrete difference equation: , where Indicates the time series Sample values, and Represent the input and output sequences of the object respectively, is zero-mean white noise, represents the independent variable of the discrete difference equation, represents an operator that shifts back one term, Indicates backward movement The operator of the term, , and are polynomials of the backshift operator of the form: , , , where , and represent the coefficients of each polynomial term, represents an operator that shifts two items backward, Indicates backward movement The operator of the term, Indicates backward movement The operator of the term, Indicates backward movement The operator of the term, express No. The coefficient of the term, express No. The coefficient of the term, express No. The coefficient of the term; First set , and assumes a set value or reference sequence , ; The following performance objective function is used for GPC optimization: , where represents the performance objective function, Because until the time The object output obtained by the input and output data The optimal prediction value is one step ahead. and are the minimum prediction time domain and the maximum prediction time domain respectively, represents the increment of input, To control the time domain, and are the output prediction deviation and control increment weighted sequence, respectively, It represents the reference trajectory of the future output of the system, which is the product of the set value after online softening; make , , ,but , , set adjustable parameters ,make , where represents the input matrix, represents the output matrix, represents the initial output matrix, and denotes parameter vectors, Indicates the number of parameters. Indicates parameters, for The matrix-vector form of .
2. The SCR denitrification and ammonia injection control method for a thermal power plant according to claim 1, It is characterized in that Assume that the transfer function of the secondary loop controller is , the transfer function of the ammonia injection system is , then the equivalent transfer function of the internal unit negative feedback control loop is : , It is obtained based on actual test data after a step disturbance experiment when the SCR denitrification and ammonia injection control system is running stably.
3. The SCR denitrification and ammonia injection control method for a thermal power plant according to claim 2, It is characterized in that Assume that the transfer function of the Smith predictor is In the distributed control system workstation, a subtraction module is used to compare the output measurement signal of the master control object with the The output signal of the master object and Apply the same ammonia injection flow input disturbance signal. If the output measurement signal of the main control object is If the output signal after subtracting the output signal of is close to zero, it is determined that the transfer function of the Smith predictor fully matches the dynamics of the actual master control object.
4. The SCR denitrification and ammonia injection control method for a thermal power plant according to claim 3, It is characterized in that The feedforward compensation algorithm corresponding to the ammonia flow intelligent feedforward controller adopts a competitive neural network model. The competitive neural network model has two layers, namely an output layer and an input layer. Only one output neuron wins each calculation. The winning output neuron is marked as 1, and the other output neurons are marked as 0. When the training of the competitive neural network model begins, the weights between the input layer and the output layer are set arbitrarily, and the min-max normalization is used for data processing. , where is the normalized data, is the minimum value of the data before normalization, is the maximum value of the data before normalization; The root mean square error (RMSE) and mean absolute percentage error (MAPE) were used to verify the accuracy of the competitive neural network model. , , where NO x The concentration of Measured values, NO x The concentration of Prediction values, is the sample size of the study.
5. The SCR denitrification ammonia injection control method for a thermal power plant according to claim 4, It is characterized in that NO in SCR Denitrification Ammonia Spray Control System x The measurement distortion disturbance generated by the concentration analyzer during maintenance is replaced by the predicted value calculated by the competitive neural network model. x Detection values of the concentration analyzer during maintenance.
