CN104826492B - Improvement method for selective catalytic reduction flue gas denitrification and ammonia injection control system - Google Patents

Abstract

The invention provides an improved method for the selective catalytic reduction flue gas denitrification ammonia injection control system, which introduces an optimization control station into the original flue gas denitrification ammonia injection control system. In the optimization control station module, the parameters affecting the NOx production of coal-fired boilers and the original PID controller output ammonia injection control signal, dynamic neural network, and the output sensitivity method are used to construct hidden layer neurons, and comprehensive learning of PID control The controller and the imported parameters are comprehensively calculated according to the output of the PID controller and the imported parameters that affect the NOx generation of the boiler to form a dynamic compensation control amount. In the control logic of ammonia injection volume for denitrification, the original PID controller is used as the main controller, and the dynamic neural network ammonia injection volume prediction controller is used as the correction controller to obtain the optimal adjustment volume of ammonia injection volume, which promotes the reliable and economical operation of the denitrification system .

Description

An improved method of selective catalytic reduction flue gas denitrification and ammonia injection control system

technical field

The invention belongs to the technical field of flue gas denitrification, and in particular relates to an improved method of an ammonia injection control system for flue gas denitrification by a selective catalytic reduction method.

Background technique

Selective catalytic reduction (SCR) flue gas denitrification technology is currently the most widely used and most effective flue gas denitrification technology in the world. The principle of this technology is that under the action of a catalyst, ammonia gas is injected to reduce the NOx in the flue gas to N ₂ and H ₂ O, and the reducing agent is mainly liquid ammonia. SCR flue gas denitrification system is mainly composed of denitrification reactor, liquid ammonia storage tank, ammonia injector, mixer, nitrogen oxide monitoring device and control system, etc., and is installed on the flue between the boiler economizer and the air preheater . During the operation of the denitrification system, the liquid ammonia is evaporated into ammonia gas by the evaporator, and after being evenly mixed with the air, according to the amount of NOx in the flue gas, the ammonia injection flow control system calculates the demand for ammonia injection, and controls the opening of the ammonia injection valve. It is sprayed from the nozzle on the ammonia injection grid pipe in front of the SCR reaction flue, and after being evenly mixed with the boiler flue gas, it flows through the catalyst layer of the SCR reactor and undergoes a chemical reaction to complete the denitrification process.

The key task of the denitrification control system is to control the amount of ammonia injected into the flue gas. The amount of ammonia injection and its control method are directly related to the denitrification efficiency of coal-fired power plant boilers, NOx emission concentration and ammonia escape rate and other indicators. Insufficient ammonia injection will lead to low denitrification efficiency, and the outlet NOx emission concentration cannot meet the requirements allowed by national regulations. Excessive ammonia injection will not only lead to higher operating costs, but also increase the production of ammonium sulfate and ammonium bisulfate. The deposition of ammonium sulfate and ammonium bisulfate will reduce the life of the catalyst and cause a series of problems such as clogging.

There are mainly two control methods for the amount of ammonia injection in the denitrification system: the fixed ammonia nitrogen molar ratio control method and the outlet NOx fixed value control method. The fixed ammonia nitrogen molar ratio control method is to use the inlet NOx concentration multiplied by the flue gas flow to obtain the NOx content signal, which is multiplied by the NH ₃ /NOx molar ratio K (between 0.7 and 1.3) to obtain the required basic ammonia injection amount, and then According to the feedback of the NOx content at the outlet, the value of ammonia to be injected is corrected, and after the operation of the PID (proportional-integral-derivative) controller, a regulating valve opening command is issued to control the flow of ammonia injection. This control method is easy to cause excessive denitrification, and it is difficult to control the escape amount of ammonia gas, which increases operating costs and secondary environmental pollution. The outlet NOx fixed value control method is to set the outlet NOx concentration value, and calculate the pre-denitration efficiency and preset molar ratio according to the actual measured value of inlet NOx. The preset molar ratio is output as the reference of the molar ratio controller. The actual measured value of outlet NOx is compared with the set value of outlet NOx, and then the output of the PID regulator is used as a correction to finally determine the molar ratio currently required by the control system. The molar ratio of this control mode is a variable value, and the ammonia nitrogen molar ratio corresponds to the NOx value at the outlet of the reactor SCR and the boiler load.

The existing ammonia injection quantity control method relies heavily on the measured value of NOx concentration. When the existing instrument tests NOx in the flue gas, the NOx analyzer feedback will be delayed for 1-2 minutes, which has a large time lag. Once the NOx in the flue gas When the concentration changes, the amount of ammonia injection cannot be adjusted in time. The amount of NOx generated in the flue gas is closely related to the type of coal burned, the load of the unit, and the operating conditions of the unit. In particular, the load change rate of existing coal-fired units is usually set relatively high. Due to the influence of grid load distribution, when the unit load fluctuates greatly (load changes of more than 50MW in a short period of time), the primary wind pressure and flue gas flow rate on the boiler side will affect the denitrification. The operating parameters of the system will also fluctuate greatly. The amount of NOx generated in the flue gas will change accordingly. The flue gas flows rapidly in the flue, and the emission concentration of NOx changes continuously. Due to the delay of the feedback signal, there is a certain deviation between the measured value of NOx and the actual value. Due to the influence of various conditions, the denitrification reaction under the action of the catalyst has a very complex reaction characteristic, which makes the SCR denitrification ammonia injection control system nonlinear. Due to the hysteresis of the measurement signal, the change of the coal type and the fluctuation of the load, the SCR denitrification system has a large hysteresis and strong nonlinearity. The traditional PID ammonia injection amount is designed according to the highest load. Once the operating condition occurs Changes will lead to a serious discrepancy between the actual ammonia injection amount and the theoretical ammonia injection amount. When the amount of ammonia injected is too small, it is difficult to guarantee the NOx emission standard; when the amount of ammonia injected is too much, it is easy to cause excess NH ₃ to escape, which not only causes waste of ammonia, but also causes secondary pollution.

Therefore, the denitrification and ammonia injection control system operates under variable working conditions. It is required to optimize the design of the control system in the SCR denitrification process to solve the nonlinearity and large hysteresis of the existing denitrification and ammonia injection control system. At the same time, the ammonia injection flow rate should be reduced as much as possible, so that the denitrification system can operate reliably and economically.

Contents of the invention

Due to the hysteresis of the NOx measurement signal in the existing flue gas, the change of the coal type and the fluctuation of the load, etc., the SCR denitrification system has a large hysteresis and strong nonlinearity. The amount of ammonia injection is seriously inconsistent with the theoretical amount of ammonia injection.

The present invention proposes to set up an optimization control station in the existing PID control system, and introduce a dynamic neural network ammonia injection quantity prediction control system into the optimization control station module to establish denitrification efficiency and inlet NOx concentration, ammonia escape amount, unit load, and combustion temperature , coal combustion, air supply volume, coal quality characteristics, excess air coefficient and other flue gas state variables to optimize the control of ammonia injection volume. The monitored boiler operation signal is introduced into the control system, and the change of ammonia injection flow caused by the change of NOx generation is predicted through model calculation, and the fluctuation phenomenon of NOx generation in coal-fired flue gas is compensated accordingly. The structure of the network is simple, the learning rate is fast, and it has good parallel computing ability and good nonlinear mapping ability, which can improve the stability and timeliness of system control. The SCR flue gas denitrification control system provided by the present invention adopts the outlet NOx fixed value control method. According to the difference between the measured outlet NOx concentration and the target outlet NOx concentration, a predictive control system module is added on the basis of the existing PID control system. In the predictive control The system module introduces the correction of flue gas state parameters such as inlet NOx concentration, ammonia escape amount, unit load, combustion temperature, coal combustion amount, and excess air coefficient, and adjusts the amount of ammonia injection to achieve the final emission concentration to meet the design requirements. The dynamic neural network predictive control system for ammonia injection includes a predictive controller and an original PID controller, among which the predictive controller is used as a correction controller. The dynamic neural network prediction controller comprehensively learns the important state parameters of flue gas, uses the sensitivity learning algorithm to determine the structure of the hidden layer, and predicts the flow rate of ammonia injection. The original PID controller is used as the main controller, and its output is introduced into the dynamic neural network ammonia injection forecast controller, and the dynamic compensation control is formed comprehensively according to the output of the main controller and other factors that affect the denitrification efficiency. The invention can overcome the non-linearity and hysteresis of the existing denitrification control system, improve the denitrification efficiency of the system and the adaptability to variable working conditions, and reduce the escape amount of ammonia gas. The dynamic neural network predictive control diagram of ammonia injection quantity is shown in Figure 1.

The invention provides an improved method for the selective catalytic reduction flue gas denitrification ammonia injection control system, which is characterized in that an optimized control station is introduced into the flue gas denitrification ammonia injection control system, and an optimized control station is introduced into the original flue gas denitrification ammonia injection control system. Optimized Control Station. In the optimized control station module, the parameters that affect the NOx production of coal-fired boilers and the original PID controller output ammonia injection control signal are introduced, and the output sensitivity method is used to construct hidden layer neurons, and the PID controller and the introduced PID controller are comprehensively studied. Parameters, based on the output of the PID controller and the introduced parameters that affect the NOx production of the boiler, are comprehensively calculated to form a dynamic compensation control quantity. In the control logic of denitrification ammonia injection volume, the original PID controller is used as the main controller, and the dynamic neural network ammonia injection volume prediction controller is used as the correction controller.

Further, the parameters affecting boiler NOx generation include denitrification reactor inlet flue gas NOx content z(t), ammonia nitrogen molar ratio η(t), combustion temperature T ₁ (t), reactor inlet temperature T ₂ (t) , reactor outlet flue gas NOx content y(t), boiler real-time load L(t), excess air coefficient α(t), coal combustion B(t), coal quality characteristics Q(t), air supply A( t), escape NH ₃ concentration C(t) and output u1(t) of the original PID controller.

Further, the denitrification ammonia injection volume control system weighted the output u1(t) of the original PID controller and the output u2(t) of the dynamic neural network prediction ammonia injection volume controller to obtain the final ammonia injection volume Measure u(t).

Further, the number of hidden layer neurons is 24.

Further, use the output sensitivity method to calculate the output contribution value ET _h of hidden layer neurons; that is, apply the following formula:

In the formula

in,

This technical solution will timely and correctly incorporate the factors affecting NOx generation into the denitrification control system, solve the delay lag problem in the NOx measurement system, effectively reduce the monitoring error caused by the delay problem, and combine the amount of ammonia injection with the concentration in the flue gas. The NOx concentration is matched in time to promote the reliable and economical operation of the denitrification system.

Description of drawings

Fig. 1 is a schematic diagram of the ammonia injection control system of the present invention.

Figure 2a is a graph of load versus time.

Figure 2b is the outlet NOx concentration under the traditional PID control mode and the optimal control station mode at different loads.

Figure 3a is a graph of load versus time.

Figure 3b shows the amount of ammonia escape under the traditional PID control mode and the optimal control station mode at different loads.

Fig. 4 is a graph showing the variation of ammonia injection amount during operation under variable working conditions.

detailed description

In order to make the purpose, technical solution and advantages of the present invention clearer, the invention will be further described in detail below in conjunction with the accompanying drawings.

In the denitrification ammonia injection control system, a dynamic neural network ammonia injection quantity predictive controller and a PID controller are used, and the predictive controller uses the output sensitivity method to construct a suitable hidden layer neuron to avoid the network structure being too large or too small , the controller comprehensively learns the original PID controller and boiler operating parameter information, overcomes the nonlinearity and time-varying nature of the system under variable load, and predicts the ammonia injection flow rate of the SCR device. This control scheme can control the amount of ammonia injection more precisely. The predictive controller first comprehensively learns the main state quantity information related to the denitrification efficiency. The goal of the network model is to minimize the deviation between the NOx content at the flue gas outlet and the set value, and to obtain the optimal ammonia injection amount. The parameters introduced by the dynamic neural network ammonia injection forecast controller include:

(1) The real-time load L(t) of the boiler, when the unit is in peak-shaving operation, affects the amount of coal burned and flue gas, which in turn affects the operating conditions of the denitrification system.

(2) The amount of coal burned B(t) and the amount of fuel input affect the amount of flue gas and NOx in flue gas.

(3) Air supply volume A(t), including primary air volume and secondary air volume, affects flue gas volume and NOx in flue gas.

(4) Coal quality characteristic Q(t), the change of coal quality will directly affect the amount of flue gas and the NOx content in flue gas.

(5) The NOx content z(t) of SCR inlet flue gas can be obtained by the continuous online monitoring system of flue gas.

(6) Ammonia-nitrogen molar ratio η(t)(NH ₃ /NOx), optimize the optimal ammonia-nitrogen ratio according to the operating conditions of the unit to achieve the set denitrification efficiency.

(7) The combustion temperature T ₁ (t), the combustion temperature of the furnace, seriously affects the amount of NOx generated in the flue gas, and affects the load and denitrification efficiency of the denitrification system.

(8) The temperature T ₂ (t) at the inlet of the reactor affects the activity of the catalyst and the catalytic reaction speed, and has a great influence on the denitrification efficiency.

(9) For the NOx content y(t) of the flue gas at the outlet of the SCR, the NOx content of the flue gas at the outlet is used to feedback and correct the control amount, which can improve the accuracy of the ammonia injection amount control.

(10) The excess air coefficient α(t), which affects the size of the flue gas volume and the formation of NOx in the coal-fired flue gas.

(11) The escape NH ₃ concentration C(t), which reflects the completeness of the denitrification reaction.

(12) The output u1(t) of the PID master controller.

Finally, the output u1(t) of the original PID controller and the output u2(t) of the dynamic neural network ammonia injection controller are weighted and summed to obtain the total output u(t), which is the adjustment amount of ammonia injection.

The input layer of the dynamic neural network predictor is:

X＝[x ₁ ,x ₂ ,x ₃ ,x ₄ ,x ₅ ,x ₆ ,x ₇ ,x ₈ ,x ₉ ,x ₁₀ ,x ₁₁ ,x ₁₂ ] ^T

=[L(t), B(t), A(t), Q(t), Z(t), η(t), T1(t), T2(t), y(t), α(t ), C(t), u2(t)] ^T

The output of the network can be expressed as:

The learning algorithm of the dynamic neural network ammonia injection forecasting controller adopts RBF network. When the network is constructed, the neuron output of the hidden layer has different influences on the output of the neural network. Split and eliminate hidden layer neurons with small contribution values to achieve the purpose of self-adjustment of hidden layer network structure. Specific steps are as follows:

1) First set the number of neurons in the hidden layer to an arbitrary natural number 24 (twice the number of neural network input), so the initial neural network structure is 12-14-1, and the neural network is preformed with a given sample group. To calculate;

2) After completing a certain number of forward calculations, find out the maximum and minimum values of the neuron output in the hidden layer;

3) Perform corresponding sensitivity analysis on the output of each neuron to obtain the output contribution value ET _h of the neuron; that is, apply the following formula:

In the formula

4) If the value of the i-th neuron in the hidden layer is the largest at time t and satisfies the ET _h >δ _{split threshold} , the neuron splits, and the width and center parameters of the initial parameters of the newly added neuron k+1 are the same as those of the original neuron k+1 The parameters of the i neurons are the same, and the connection weight w _k+1 = rand(0,1)×w _i , w _i = _wi -w _K+1 ; if the jth neuron in the hidden layer at time t value is the smallest, and satisfy the ET _h <δ _{deletion threshold} , then record the neuron with the closest Euclidean distance as jj, delete the jth neuron, and the connection weight of the jjth neuron becomes

5) According to the error function, the weight, center value and center width of neurons in the hidden layer are adjusted by gradient descent method;

in

6) Stop the calculation when the desired error or calculation step is reached.

When the neural network is initially trained, the output has nothing to do with the controlled characteristics, so its influence on the output of the main control variable is 0. After a period of sufficient training, according to the mapping structure of the neural network, the output of the neural network has established The nonlinear mapping relationship between the input value of the neural network (including the parameters affecting NOx and the output of the PID controller) and the optimal ammonia injection amount, and the output of the predictive controller is the corresponding optimal ammonia injection amount under the working conditions of each input parameter , so it can make its influence gradually increase. When there are drastic changes in system load, large changes in coal quality characteristics and other factors that affect the change of flue gas NOx concentration and the measurement accuracy of NOx concentration, the PID main controller should be set to The output of the system becomes smaller, and the main function of the system at this time is the dynamic predictive controller. It can be seen from the system diagram in Figure 1 that the ammonia injection control quantity u(t) is

u(t)=u1(t)+u2(t)=u _PID ×K+u _prediction ×(1-K)

Example 1

The simulation experiment results of a 600MW coal-fired boiler denitrification system in variable load operation, where Figure 2a is the boiler load change curve with time, and Figure 2b is the NOx emission concentration at the outlet of the denitrification system. The denitrification control system operates in the outlet NOx fixed value control mode, and the set outlet NOx concentration is 50ppm. It operates in two control modes: the traditional PID control mode and the optimal control mode of the present invention. It can be seen from Fig. 2b that when the unit is operating with variable load, the NOx emission concentration at the outlet of the denitrification system increases when the traditional PID control method is adopted, and will exceed the set value; The probability of the NOx concentration exceeding the set value is relatively low, and the NOx concentration at the outlet of the denitrification system is lower than that of the traditional PID control method.

Example 2

The simulation experiment results of a 600MW coal-fired boiler denitrification system in variable load operation, where Figure 3a is the boiler load change curve with time, and Figure 3b is the amount of ammonia escape at the outlet of the denitrification system. The denitrification control system operates in the outlet NOx fixed value control mode, and the set outlet NOx concentration is 50ppm. It operates in two control modes: the traditional PID control mode and the optimal control mode of the present invention. It can be seen from Fig. 3b that the denitrification operation control provided by the present invention can reduce the amount of ammonia escape from the denitrification system. The reason is that the control system provided by the present invention comprehensively considers various parameters affecting the denitrification efficiency, and can accurately control the amount of ammonia injection in real time, which can reduce the escape amount of ammonia gas.

Example 3

Fig. 4 is a comparison of the amount of ammonia sprayed by the denitrification system of a 600MW coal-fired boiler denitrification system when the traditional PID control mode and the control mode provided by the present invention are respectively adopted during variable load operation. It can be seen from Figure 4 that when the SCR denitrification system is operated in the optimized control station control mode, the amount of ammonia injection can be saved when the outlet NOx concentration standard is met.

Claims (1)

1. An improved method of a selective catalytic reduction flue gas denitration ammonia injection control system is characterized in that: introducing an optimization control station into the original flue gas denitration ammonia injection control system; in an optimization control station module, introducing parameters influencing the NOx generation amount of the coal-fired boiler and an ammonia injection amount control signal output by an original PID controller, constructing a hidden layer neuron by using an output sensitivity method, comprehensively learning the PID controller and the introduced parameters, and comprehensively calculating to form dynamic compensation control amount according to the output amount of the PID controller and the introduced parameters influencing the NOx generation amount of the boiler; in the denitration ammonia injection amount control logic, an original PID controller is used as a main controller, and a dynamic neural network ammonia injection amount prediction controller is used as a correction controller;

the parameters introduced by the dynamic neural network ammonia injection amount prediction controller comprise:

(1) the real-time load L (t) of the boiler influences the coal burning quantity and the smoke quantity when the unit is in peak shaving operation, and further influences the operation condition of the denitration system;

(2) the coal burning quantity B (t) and the fuel input quantity influence the smoke quantity and NOx in the smoke;

(3) the air supply quantity A (t) comprises primary air and secondary air quantity, and influences the quantity of flue gas and NOx in the flue gas;

(4) the coal quality characteristic Q (t), the change of the coal quality can directly influence the amount of the flue gas and the content of NOx in the flue gas;

(5) the SCR inlet flue gas NOx content z (t) can be obtained by a flue gas continuous online monitoring system;

(6) ammonia nitrogen molar ratio η (t) (NH)₃NOx), the optimal ammonia nitrogen ratio is optimized according to the unit operation conditions to realize the set denitration efficiency;

(7) combustion temperature T₁(t), the combustion temperature of a hearth seriously influences the generation amount of NOx in the flue gas, and influences the load of a denitration system and the denitration efficiency;

(8) reactor inlet temperature T₂(t), the activity and the catalytic reaction speed of the catalyst are influenced, and the denitration efficiency is greatly influenced;

(9) the NOx content y (t) of the flue gas at the outlet of the SCR adopts the NOx content feedback correction control quantity of the flue gas at the outlet, so that the control accuracy of the ammonia injection quantity can be improved;

(10) the excess air coefficient alpha (t) influences the size of the flue gas quantity and the generation of NOx in the coal-fired flue gas;

(11) slip NH₃Concentration c (t) reflecting the degree of completion of the denitration reaction;

(12) PID main controller output u1 (t);

finally, weighting and summing the output quantity u1(t) of the original PID controller and the output quantity u2(t) of the dynamic neural network ammonia injection quantity controller to obtain a total output quantity u (t), namely the regulating quantity of the ammonia injection quantity;

the input layers of the dynamic neural network predictor are as follows:

X＝[x₁，x₂，x₃，x₄，x₅，x₆，x₇，x₈，x₉，x₁₀]^T＝[L(t)，B(t)，A(t)，Q(t)，Z(t)，η(t)，T1(t)，T2(t)，y(t)，α(t)，C(t)，u2(t)]^τ

the output of the network can be expressed as:

the dynamic neural network predictor learning algorithm adopts RBF network, when the network is constructed, the influence degree of the neuron output of the hidden layer on the neural network output is different, so that the neuron of the hidden layer with large influence on the output is split, the neuron of the hidden layer with small contribution value is eliminated, and the purpose of self-adjustment of the hidden layer network structure is achieved, and the method specifically comprises the following steps:

1) firstly, setting the number of neurons in a hidden layer to be 24 (2 times of the number of input quantity of the neural network) of any natural number, so that the structure of the initial neural network is 12-14-1, and performing forward calculation on the neural network by using a given sample group;

2) after forward calculation for a certain number of times is completed, finding out the maximum value and the minimum value of the hidden layer neuron output;

3) carrying out corresponding sensitivity analysis on the output of each neuron to obtain the output contribution value ET of the neuron_h(ii) a Namely, the following formula is applied:

in the formula

4) If the ith neuron in the hidden layer has the maximum value at the time t and satisfies ET_h>_{Splitting threshold}Splitting the neuron, wherein the width and the central parameter in the initial parameter of the newly added neuron k +1 are the same as the parameters of the original ith neuron, and connecting the weight w_k+1＝rand(0,1)×w_i,w_i＝w_i-w_K+1(ii) a If the value of the jth neuron in the hidden layer is minimum at the time t and satisfies ET_h<_{Threshold value of deletion}Recording the nearest neuron in Euclidean distance as jj, deleting the jth neuron, and changing the connection weight of the jth neuron into

5) Adopting a gradient descent method to adjust the weight, the central value and the central width of the hidden layer neuron according to the error function;

wherein

6) Stopping the calculation when the expected error is reached or the calculation step is completed;

when the neural network is initially trained, the output is irrelevant to the controlled characteristic, so the output influence effect of the neural network on the main control quantity is 0, after a period of time of full training, according to the mapping structure of the neural network, the output of the neural network establishes the nonlinear mapping relation between the input value of the neural network and the optimal ammonia injection quantity, the output of the prediction controller is the optimal ammonia injection quantity value corresponding to each input parameter under the working condition, so the influence effect can be gradually increased, when the system load is changed violently, the coal quality characteristic is changed greatly and other factors influencing the change of the smoke NOx concentration and influencing the measurement precision factor of the measured NOx concentration, the output quantity of the PID controller is reduced, and the system plays the main role of a dynamic prediction controller at the moment.