CN110026068A - A kind of large-scale coal fired power plant CO based on Neural network inverse control2Trapping system and feed forward control method - Google Patents

Abstract

The invention discloses a large-scale coal-fired power station CO ₂ capture system and a feedforward control method based on neural network inverse control. The coal-fired power station CO ₂ capture system is regarded as a five-input-five-output multivariable system, and the main Steam pressure, outlet enthalpy value of steam-water separator, unit power generation, CO ₂ capture rate and reboiler temperature are the main controlled variables, and the coal supply, water supply, main steam valve, lean liquid flow and reboiler are selected. The steam flow is the corresponding control variable. The invention adopts the BP neural network technology to establish the inverse model of the CO ₂ capture system of the large coal-fired power station, so that the required control variables can be calculated according to the given values, so as to realize the advance control, and can effectively deal with the large delay characteristics of the overall system and improve the The dynamic adjustment quality of the output side; in addition, the correction of the neural network inverse model is realized by adding a PID control compensator, thereby enhancing its anti-disturbance and uncertainty ability, making the control system adapt to the needs of industrial sites.

Description

A CO2 capture system and feedforward for large coal-fired power station based on neural network inverse control Control Method

technical field

The invention relates to the field of thermal automatic control, in particular to a large-scale coal-fired power station _CO2 capture system and a feedforward control method based on neural network inverse control.

Background technique

Thermal power plants are currently the most important emission source of CO ₂ gas, which has a great impact on the greenhouse effect. Post-combustion CO ₂ capture technology based on chemisorption is an important measure to achieve CO ₂ capture and reduce greenhouse gas emissions. The post-combustion CO ₂ capture technology using MEA as the adsorption solvent has become the mainstream of commercial CO ₂ capture technology in the world due to its high efficiency, high economy, mature technology and easy adjustment. At the same time, post-combustion CO ₂ The capture technology does not need to change the operation structure of the existing thermal power unit, and the capture equipment can be effectively operated after the tail flue, which reduces the investment cost.

Thermal power units have strong coupling characteristics with post-combustion _CO2 capture systems. According to the power grid load command, the thermal power unit needs to participate in load peak regulation, so the tail flue gas will fluctuate randomly with the group load, and the flue gas fluctuation will then affect the downstream CO ₂ capture system, and affect key variables such as capture rate and reboiler temperature. On the other hand, the reboiler steam in the post-combustion CO ₂ capture system is provided by the exhaust gas of the steam turbine, which will reduce the power generation of the unit and affect the peak regulation of the unit. Considering the coupling characteristics between the thermal power unit and the capture system, it is necessary to comprehensively consider the two and optimize the control as a whole system. At the same time, studies have shown that the large-scale coal-fired power station capture system has large inertia and delay, and the existence of disturbance, measurement noise, and uncertainty will also interfere with the controller to a certain extent, and it is difficult to obtain good control quality. At present, the conventional PID control scheme is usually adopted for the CO ₂ capture system of large coal-fired power plants, which is difficult to effectively deal with the large delay and strong coupling characteristics of the controlled object.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide a large-scale coal-fired power station CO ₂ capture system and a feedforward control method based on neural network inverse control, which can reduce the dynamic deviation caused by large inertia, perform control in advance, and improve control quality .

In order to solve the above technical problems, the present invention provides a large-scale coal-fired power station CO ₂ capture system based on neural network inverse control, comprising: a target value setting unit 1, a neural network inverse controller 2, a PID control compensator 3, a large-scale combustion Coal power station _CO2 capture overall system model 4, first delay unit 5 and second delay unit 6; target value setting unit 1 has two outputs, which are respectively connected with neural network inverse controller 2 and PID compensation controller 3; The deviation e(k) between the output r(k+1) of the target value setting unit 1 and the output y(k+1) of the large-scale coal-fired power station _CO2 capture overall system model 4 is used as the input of the PID control compensator 3 to solve Compensation input variable u _PID (k); the input u (k) of the overall system model 4 for CO ₂ capture in large coal-fired power plants is the output u _PID (k) of the PID control compensator 3 and the output u _NN of the neural network inverse controller 2 The sum of (k); the input variable u(k) and the output variable y(k+1) of the large coal-fired power station _CO2 capture overall system model 4 pass through the first delay unit 5 and the second delay unit 6, respectively, to obtain a delay Variables u(k-1) and y(k); the first delay unit 5 and the second delay unit 6 output variables u(k-1) and y(k) and the target value setting unit 1 outputs r(k+1 ) as the input of the neural network inverse controller 2, and the output u _NN (k) is calculated.

Correspondingly, a feedforward control method for a large-scale coal-fired power station CO ₂ capture system based on neural network inverse control includes the following steps:

(1) Select the main steam pressure, the outlet enthalpy of the steam-water separator, the power generation of the unit, the CO ₂ capture rate and the temperature of the reboiler as the controlled variables of the CO ₂ capture system model 4 of the large coal-fired power station, and select the coal feed of the unit volume, feed water volume, main steam valve, lean liquid flow and reboiler steam flow are the corresponding control variables;

(2) In the case of closed-loop, change the given values of the controlled variables such as flue gas and capture rate, and conduct a closed-loop response test; set the sampling period T to obtain the CO ₂ capture of large coal-fired power plants under different flue gas and capture rate loads. Set the steady-state and dynamic parameters of the control variable and the controlled variable of the system model 4;

(3) Take the controlled quantity data of the _CO2 capture system model 4 of the large-scale coal-fired power station as the output, and use the controlled quantity data of the _CO2 capture system model 4 of the large-scale coal-fired power station as the input, and use the BP neural network for offline training, Determine the inverse system model of the _CO2 capture system model 4 of the large coal-fired power station, such as formula (1):

u _NN (k)=f(y(k+1),y(k),…,y(kn ₁ ),u(k-1),…,u(kn ₂ )) (1)

(4) Set up a control loop to control the main steam pressure with the coal feed of the unit, control the outlet enthalpy of the steam-water separator with the water feed, control the power generation of the unit with the main steam valve, control the CO ₂ capture rate with the lean liquid flow, and use the recycled Boiler steam flow controls reboiler temperature;

(5) Setting the relevant parameters of the PID control compensator 3, including proportional gain k _P , integral time constant T _i , differential gain k _d , and differential time constant T _d ;

(6) Take the output r(k+1) of the target value setting unit 1 and the outputs u(k-1) and y(k) of the first delay unit 5 and the second delay unit 6 as input variables respectively, and use the formula ( 1) Calculate the output u _NN (k) of the neural network inverse controller 2 at time k;

(7) Compare the output r(k+1) of the target value setting unit 1 with the output y(k+1) of the _CO2 capture system model 4 of the large coal-fired power station, and calculate the output error e(k); use the output error As the input of the PID control compensator 3, the compensation input u _PID (k) is calculated; using formula (2):

(8) Calculate the actual output of the _CO2 capture system model 4 of the large coal-fired power station at time k; use formula (3):

u(k)= _uNN (k)+ _uPID (k)(3)

(9) Steps (6) to (8) are repeatedly executed in the subsequent cycles to obtain the corresponding control amount and realize the error-free control.

Preferably, in step (2), the selection rule of sampling time T is T ₉₅ /T=5-15, wherein T ₉₅ is the adjustment time for the unit step response process of the object to rise to 95%.

Preferably, in step (5), the selection rule of proportional gain k _P , integral time constant T _i , differential gain k _d , and differential time constant T _d is the Ziegler-Nichols engineering tuning method.

The beneficial effects of the present invention are as follows: the present invention can improve the quality of dynamic adjustment by adopting the feedforward control method based on neural network inverse control; meanwhile, by introducing a PID compensation controller, it can effectively deal with the influences caused by the mismatch of the prediction model and the disturbance, etc. So as to ensure the control quality of the joint supply system.

Description of drawings

FIG. 1 is a schematic structural diagram of a control system of the present invention.

Fig. 2 is a schematic flow chart of the _CO2 capture system of the large coal-fired power station of the present invention.

Fig. 3(a) is a schematic diagram showing the comparison of the main steam pressure control effect on the output side of the coal-fired unit when the present invention and the traditional PID controller change in a stepwise change of the given value.

Figure 3(b) is a schematic diagram showing the comparison of the enthalpy value control effect at the outlet of the steam-water separator at the output side of the coal-fired unit when the present invention and the traditional PID controller change in a step change of the given value.

Figure 3(c) is a schematic diagram showing the comparison of the power generation control effect of the coal-fired unit output side unit when the given value is step-changed by the present invention and the traditional PID controller.

Figure 4(a) is a schematic diagram showing the comparison between the present invention and the traditional PID controller when the given value is changed in a step-by-step manner.

Figure 4(b) is a schematic diagram showing the comparison between the present invention and the traditional PID controller in the control effect of the input side water supply of the coal-fired unit when the given value is changed in a stepwise manner.

Figure 4(c) is a schematic diagram of the comparison between the present invention and the traditional PID controller when the given value is changed in a step-by-step manner.

Figure 5(a) is a schematic diagram showing the comparison of the CO ₂ capture rate control effect on the output side of the CO ₂ capture system between the present invention and the traditional PID controller when the given value is changed in a stepwise manner.

Figure 5(b) is a schematic diagram showing the comparison of the temperature control effect of the reboiler at the output side of the _CO2 capture system when the present invention and the traditional PID controller change in a step change of the given value.

Figure 6(a) is a schematic diagram showing the comparison of the lean liquid flow control effect on the input side of the CO ₂ capture system when the present invention and the traditional PID controller change in a step change of the given value.

Figure 6(b) is a schematic diagram showing the comparison of the steam flow control effect of the reboiler on the input side of the CO ₂ capture system when the present invention and the traditional PID controller change in a step change of the given value.

Detailed ways

As shown in Figure 1, a large-scale coal-fired power station _CO capture system based on neural network inverse control includes: target value setting unit 1, neural network inverse controller 2, PID control compensator 3, large-scale coal-fired power station CO ₂ Capture the overall system model 4, the first delay unit 5 and the second delay unit 6; the target value setting unit 1 has two outputs, which are respectively connected with the neural network inverse controller 2 and the PID compensation controller 3; the target value is set The deviation e(k) between the output r(k+1) of the fixed unit 1 and the output y(k+1) of the large-scale coal-fired power station _CO2 capture overall system model 4 is used as the input of the PID control compensator 3, and the compensation input variable is solved. u _PID (k); the input u (k) of the overall system model 4 for CO ₂ capture in large coal-fired power plants is the output of PID control compensator 3 u _PID (k) and the output of neural network inverse controller 2 u _NN (k) The sum; the input variable u(k) and output variable y(k+1) of the _CO2 capture overall system model 4 of the large-scale coal-fired power station pass through the first delay unit 5 and the second delay unit 6, respectively, to obtain the delay variable u( k-1) and y(k); the first delay unit 5 and the second delay unit 6 output variables u(k-1) and y(k) and the target value setting unit 1 outputs r(k+1) as the nerve The network inverse controller 2 inputs and calculates the output u _NN (k).

As shown in Figure 2, the CO ₂ capture system of a large coal-fired power station includes: main steam pressure, outlet enthalpy value of steam-water separator, unit power generation, CO ₂ capture rate, reboiler temperature and unit coal feeding amount, feeding capacity Key variables such as water volume, main steam valve, lean liquid flow and reboiler steam flow. A feedforward control method for a large-scale coal-fired power station CO ₂ capture system based on neural network inverse control, comprising the following steps:

(1) Select the main steam pressure, the outlet enthalpy of the steam-water separator, the power generation of the unit, the CO ₂ capture rate and the temperature of the reboiler as the controlled variables of the CO ₂ capture system model 4 of the large coal-fired power station, and select the coal feed of the unit volume, feed water volume, main steam valve, lean liquid flow and reboiler steam flow are the corresponding control variables;

(2) In the case of closed-loop, change the given values of the controlled variables such as flue gas and capture rate, and conduct a closed-loop response test; set the sampling period T to obtain the CO ₂ capture of large coal-fired power plants under different flue gas and capture rate loads. Set the steady-state and dynamic parameters of the control variable and the controlled variable of the system model 4;

(3) Take the controlled quantity data of the _CO2 capture system model 4 of the large-scale coal-fired power station as the output, and use the controlled quantity data of the _CO2 capture system model 4 of the large-scale coal-fired power station as the input, and use the BP neural network for offline training, Determine the inverse system model of the _CO2 capture system model 4 of the large coal-fired power station, such as formula (1):

u _NN (k)=f(y(k+1),y(k),…,y(kn ₁ ),u(k-1),…,u(kn ₂ )) (1)

(4) Set up a control loop to control the main steam pressure with the coal feed of the unit, control the outlet enthalpy of the steam-water separator with the water feed, control the power generation of the unit with the main steam valve, control the CO ₂ capture rate with the lean liquid flow, and use the recycled Boiler steam flow controls reboiler temperature;

(5) Setting the relevant parameters of the PID control compensator 3, including proportional gain k _P , integral time constant T _i , differential gain k _d , and differential time constant T _d ;

(6) Take the output r(k+1) of the target value setting unit 1 and the outputs u(k-1) and y(k) of the first delay unit 5 and the second delay unit 6 as input variables respectively, and use the formula ( 1) Calculate the output u _NN (k) of the neural network inverse controller 2 at time k;

(7) Compare the output r(k+1) of the target value setting unit 1 with the output y(k+1) of the _CO2 capture system model 4 of the large coal-fired power station, and calculate the output error e(k); use the output error As the input of the PID control compensator 3, the compensation input u _PID (k) is calculated; using formula (2):

(8) Calculate the actual output of the _CO2 capture system model 4 of the large coal-fired power station at time k; use formula (3):

u(k)= _uNN (k)+ _uPID (k)(3)

(9) Steps (6) to (8) are repeatedly executed in the subsequent cycles to obtain the corresponding control amount and realize the error-free control.

Example:

(1) Determine the control loop and the corresponding control and controlled quantities of the _CO2 capture system of the large-scale coal-fired power station, as shown in Table 1:

Table 1

(2) Set the sampling time T=30s, use the controlled quantity data as the neural network input, the controller data as the neural network data, and use the BP neural network toolbox to establish the inverse model of the CO ₂ capture system of the coal-fired power station. The neural network contains two hidden layers, the number of neurons is 20 and 5 respectively, and the training function is trainingdm;

(3) According to the given value r(k+1) and the past input data u(k-1) and output data y(k), calculate the neural network inverse controller output u _NN (k);

(4) Set the relevant parameters of the PID control compensator, as shown in formula (4):

(5) Calculate the deviation. e(k)=r(k+1)-y(k+1);

(6) Calculate the PID control compensation output according to the deviation e(k) and formula (5):

(7) Calculate the coal supply, water supply, main steam valve, lean liquid flow and reboiler steam flow u(k) = u _NN (k) + u _PID (k) at the next moment;

(8) Output the optimal control quantity u(k), and calculate and update the neural network inverse input u _NN (k) at the next moment according to the measurement signal. Thereafter, in each sampling period, steps (3) to (8) are repeated.

The comparison between the control effect of the feedforward control method based on the neural network inverse control of the present invention and the traditional PID control effect is shown in FIG. 3(a)-FIG. 6(b). The initial steady state conditions are u ₁ =60.4620kg/s, u ₂ =425.2630kg/s, u ₃ =92.31%, u ₄ =513.4947kg/s, u ₅ =135.874kg/s, y ₁ =21.3693MPa , y ₂ =2722.1325kJ/kg, y ₃ =432.9270MWe, y ₄ =90%, y ₅ =392.2k, at 600 seconds, the output target value changes to 24.8430MPa, 2674.4886kJ/kg, 540MWe, 90 respectively %, 392.2k, after running for a period of time, the output target value changes to 24.01MPa, 2702.4781kJ/kg, 506.896MWe, 90%, 392.2k at 10500 seconds. The system runs for a total of 20,400 seconds. For the convenience of observation and comparison, the sampling period is 30 seconds for point sampling and drawing. As shown in Fig. 3(a)-Fig. 6(b), the control effect of the feedforward controller based on neural network inverse control is better, the fluctuation is small, and the response speed is fast; There is no deviation from the given value.

The invention takes the _CO2 capture system of a large coal-fired power station as a multi-variable object with five inputs and five outputs, adopts the feedforward control technology based on neural network inverse control, and selects the coal supply, water supply, main steam valve and lean liquid of the unit. Flow and reboiler steam flow are control variables, which control main steam pressure, steam-water separator outlet enthalpy, unit power generation, CO ₂ capture rate and reboiler temperature respectively. On the one hand, it can predict the system input and control it in advance, which can effectively deal with the large delay characteristics of the overall system; in addition, by introducing a PID compensation controller, it can effectively deal with the effects of prediction model adaptation, disturbance, etc., so as to ensure the overall system. Control quality.

By using the feedforward control method based on the neural network inverse controller, the present invention can predict the control variables required by the overall system in advance, better realize the coordinated control of the overall system, and improve the dynamic adjustment quality of the system; at the same time, by adding PID The compensation controller corrects the error of the neural network pseudo-model to achieve error-free control and increase the system's ability to resist external disturbances and uncertain disturbances, so that the control system can better adapt to the industrial site and improve the control quality.

Claims (4)

1. a kind of large-scale coal fired power plant CO based on Neural network inverse control₂Trapping system characterized by comprising target value is set Set unit (1), nerve network reverse controller (2), PID control compensator (3), large-scale coal fired power plant CO₂Trap total system mould Type (4), the first delay cell (5) and the second delay cell (6)；Target value setup unit (1) has two-way output, respectively with nerve Network inverse controller (2) is connected with PID compensating controller (3)；Target value setup unit (1) exports r (k+1) and large-scale coal-fired electricity Stand CO₂Input of the deviation e (k) of total system model (4) output y (k+1) as PID control compensator (3) is trapped, is solved Compensate input variable u_PID(k)；Large-scale coal fired power plant CO₂The input quantity u (k) of total system model (4) is trapped as PID control benefit Repay device (3) output u_PID(k) u is exported with nerve network reverse controller (2)_NNThe sum of (k)；Large-scale coal fired power plant CO₂The whole system of trapping The input variable u (k) and output variable y (k+1) of system model (4) pass through the first delay cell (5) and the second delay cell respectively (6), lagged variable u (k-1) and y (k) are obtained；First delay cell (5) and the second delay cell (6) output variable u (k-1) with Y (k) and target value setup unit (1) output r (k+1) are inputted as nerve network reverse controller (2), calculate output u_NN(k)。

2. a kind of large-scale coal fired power plant CO based on Neural network inverse control₂Trapping system feed forward control method, which is characterized in that Include the following steps:

(1) main steam pressure is chosen, steam-water separator exports enthalpy, unit generation amount, CO₂Capture rate and reboiler temperature are big Type coal fired power plant CO₂The controlled variable of trapping system model (4) chooses unit coal-supplying amount, confluent, main steam valve, lean solution Flow and reboiler steam flow are corresponding control variable；

(2) under closed-loop case, change flue gas, capture rate controlled variable given value, carry out closed loop response test；Setting sampling week Phase T obtains large-scale coal fired power plant CO under different flue gases, capture rate load₂The control amount of trapping system model (4) and controlled volume Stable state, dynamic parameter；

(3) by large-scale coal fired power plant CO₂The control measure of trapping system model (4) is according to as output, by large-scale coal fired power plant CO₂ The controlled volume data of trapping system model (4) carry out off-line training as input, using BP neural network, determine large-scale coal-fired electricity Stand CO₂The inverse system model of trapping system model (4), such as formula (1):

u_NN(k)=f (y (k+1), y (k) ..., y (k-n₁),u(k-1),…,u(k-n₂)) (1)

(4) control loop is set, is exported using unit Limestone control main steam pressure, using confluent control steam-water separator Enthalpy controls unit generation amount using main steam valve door, utilizes lean solution flow control CO₂Capture rate utilizes reboiler steam stream Amount control reboiler temperature；

(5) relevant parameter of PID control compensator (3), including proportional gain k are set_P, integration time constant T_i, the differential gain k_d, derivative time constant T_d；

(6) by target value setup unit (1) output r (k+1) and the first delay cell (5), the output u of the second delay cell (6) (k-1) the output u of k moment nerve network reverse controller (2) is calculated using formula (1) respectively as input variable with y (k)_NN (k)；

(7) by target value setup unit (1) output r (k+1) and large-scale coal fired power plant CO₂Trapping system model (4) exports y (k+1) It is compared, calculates output error e (k)；Output error is used as the input of PID control compensator (3), calculates compensation input Measure u_PID(k)；Using formula (2):

(8) k moment large size coal fired power plant CO is calculated₂Trapping system model (4) reality output；Using formula (3):

U (k)=u_NN(k)+u_PID(k) (3)

(9) step (6) is executed repeatedly in the period later to step (8), is obtained corresponding control amount, is realized indifference control.

3. the large-scale coal fired power plant CO based on Neural network inverse control as claimed in claim 2₂Trapping system feedforward control side Method, which is characterized in that in step (2), the selection rule of sampling time T is T₉₅/ T=5~15, wherein T₉₅For the unit of object Step response process rises to 95% regulating time.

4. the large-scale coal fired power plant CO based on Neural network inverse control as claimed in claim 2₂Trapping system feedforward control side Method, which is characterized in that in step (5), proportional gain k_P, integration time constant T_i, differential gain k_d, derivative time constant T_d's Selection rule is Ziegler-Nichols practical tuning method.