Intelligent coordinated controller design for a 600 MW supercritical boiler unit based on expanded-structure neural network inverse models

Control Engineering Practice ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at ScienceDirect

Control Engineering Practice journal homepage: www.elsevier.com/locate/conengprac

Intelligent coordinated controller design for a 600 MW supercritical boiler unit based on expanded-structure neural network inverse models$ Liangyu Ma a,n, Kwang Y. Lee b, Zhiyan Wang c a

Department of Automation, North China Electric Power University, Baoding, Hebei 071003, PR China Department of Electrical and Computer Engineering, Baylor University, Waco, TX 76798-7356, USA c Shougang Jingtang United Iron & Steel Co. Ltd., Tangshan, Hebei 063200, PR China b

art ic l e i nf o

a b s t r a c t

Article history: Received 6 May 2015 Received in revised form 25 July 2015 Accepted 2 September 2015

Under present widespread automatic generation control (AGC) centered on regional power grid, a largecapacity coal-fired supercritical (SC) power unit often operates under wide-range variable load conditions. Since a SC once-through boiler unit is represented by a typical multivariable system with large inertia and non-linear, slow time-variant and time-delay characteristics, it often makes the coordinated control quality deteriorate under wide-range loading conditions, and thus influences the unit load response speed and leads to heavy fluctuation of the main steam pressure. To improve the SC unit’s coordinated control quality with advanced intelligent control strategy, the neural-network (NN) based expanded-structure inverse system models of a 600 MW SC boiler unit were investigated. A feedforward neural network with time-delayed inputs and time-delayed output feedbacks was adopted to establish the inverse models for the load and the main steam pressure characteristics. Based on the model, a neural network inverse coordinated control scheme was designed and tested in a full-scope power plant simulator of the given SC power unit, which showed that the proposed coordinated control scheme can achieve better control results compared to the original PID coordinated control. & 2015 Elsevier Ltd. All rights reserved.

Keywords: Supercritical power unit Artificial neural network Inverse system model Coordinated control system Intelligent controller design

1. Introduction Supercritical (SC) and ultra-supercritical (USC) power generating units have become the dominant coal-fired power units in China and around the world (Garduno & Lee, 2005; Ma & Lee, 2011). With the widespread implementation of Automatic Generation Control (AGC) centered on power supply quality in a regional power grid, these large-capacity energy conversion units are required to participate in peaking-load regulation frequently without exception and often operates under large-scale variable load conditions (Gerhard, 1998; Jan, Tommy, Palle, & Tom, 1998; Garduno & Lee, 2000, 2002; Li & Wang, 2005; Jia, Cheng, & Xiong, 2007; Li et al., 2010; Wang & Li, 2011). Since a SC/USC boiler unit can be described as a strongly coupled nonlinear multivariable system with large time-delay characteristics, the traditional coordinated control strategy cannot well ☆ The short version of this paper was presented in IFAC World Congress 2014 entitled as “Neural network inverse control for the coordinated system of a 600 MW supercritical boiler unit”. n Corresponding author. Fax: þ 86 312 752 2164. E-mail addresses: [email protected] (L. Ma), [email protected] (K.Y. Lee), [email protected] (Z. Wang).

adapt to the load regulation, and often leads to slow load response and large main steam pressure fluctuations. Therefore, considering the stability of the power grid and the safety and economy of the power unit, it is of great importance to improve the coordinated control quality of the SC/USC power unit with advanced modelbased intelligent control strategies, such as neural network inverse control or predictive optimal control method (Han, Zhang, & Zhang, 2001; Heo & Lee, 2008; Lee, Helo, Hoffman, & Kim, 2007b; Lee, Ma, & Boo, 2009; Lee, Van Sickel, Hoffman, Jung, & Kim 2010; Ma & Lee, 2011, Ma, Shi, & Lee, 2010; Ma, Lee, & Ge, 2012). With increasing number of SC/USC boiler units, many studies have been performed for modelling (Ding, Li, & Wang, 2011; Yan, Zeng, Liu, & Liang, 2012). Among them, parameter tuning for some models is complicated and the models are inaccurate. Some models are too complex to fit for intelligent coordinated controller design. Therefore, how to establish a nonlinear mathematical model with higher accuracy and simpler structure that is suitable for intelligent controller design and applicable for a SC/USC boiler unit remains an important open problem. Self-adaptive inverse control method was firstly proposed and developed by Widrow and others (Widrow, McCool, & Medoff, 1978; Widrow, 1986; Widrow & Walach, 1996). It has drawn much

http://dx.doi.org/10.1016/j.conengprac.2015.09.002 0967-0661/& 2015 Elsevier Ltd. All rights reserved.

Please cite this article as: Ma, L., et al. Intelligent coordinated controller design for a 600 MW supercritical boiler unit.... Control Engineering Practice (2015), http://dx.doi.org/10.1016/j.conengprac.2015.09.002i

2

L. Ma et al. / Control Engineering Practice ∎ (∎∎∎∎) ∎∎∎–∎∎∎

attention in engineering applications with its advantages of clear physical concept, being intuitive and easy to understand (Yuan & Guo, 1994; Han et al., 2001; Dai, He, Zhang, & Zhang, 2003; Dai, 2005; Xie, Zhang, & Xiao, 2007). But solving for the inverse system model of a complex multivariable system is a bottleneck. At the same time, artificial neural networks (ANNs) have been widely used for modelling and control of complex industrial dynamic systems with impressive identification ability, strong fault tolerance and adaptive learning capability (Gencay & Liu, 1997; Lee, Heo, Hoffman, & Kim, 2007a). Combining them together, ANN-model based inverse control method can overcome the difficulty of solving the inverse problem, and present promising future applications. Recently, neural network inverse control has been applied in different areas, including power plant steam temperature control and optimization (Malinowski, Zurada, & Lilly, 1995; Wang, Yu, & Song, 2002; Dai et al., 2003; Dai, 2005; Lee et al., 2009; Ma & Lee, 2011). Aimed at improving the coordinated control quality of a largescale coal-fired power unit, the neural network inverse system models for a 600 MW SC boiler unit are studied. Two separate inverse models for the load and main steam pressure are constructed. The inputs and outputs of each model are determined by analyzing the correlation between input and output variables, and the coordinated control modes of the SC power generating unit. The models are built with time-delayed feedforward neural networks, trained and verified with abundant operating data. Based on the developed models, neural network inverse controllers are designed for the coordinated system of the 600 MW SC power unit, and the controllers are validated by real-time control simulation te1sts (Ma, Wang, & Lee, 2014).

2. Coordinated control modes and simplified model of a SC boiler unit 2.1. Coordinated control modes for a SC boiler unit Usually, the coordinated control system of a SC/USC boiler unit includes boiler master control (BMC), turbine master control (TMC), target load and load ramping-rate setting, target pressure and pressure rate setting, primary frequency tuning and other function loops (Zhang, Yu, & Song, 2007). For the 600 MW SC boiler unit investigated in this work, based on whether BMC and TMC are put into automatic or not, there are 4 kinds of control modes: (1) manual mode (both BMC and TMC are in manual), (2) boiler-following mode (TMC is in manual while BMC is in automatic), (3) turbine-following mode (BMC is in manual while TMC is in automatic), and (4) coordinated control mode (both BMC and TMC are in automatic modes). According to the inner logic difference, the coordinated control mode can be divided into Boiler-Following Based Coordinated Control (BFCC) mode and Turbine-Following Based Coordinated Control (TFCC) mode. Under BFCC mode, TMC is used to control the load by changing the valve opening of the turbine governor when load demand changes, and BMC is responsible for maintaining the main steam pressure by changing the fuel flow. It results in faster load response and smaller load deviation, but relatively larger main steam pressure fluctuations. Under TFCC mode, BMC is responsible for controlling the load when load demand changes, and TMC is used to maintain the pressure by changing the turbine valve opening. It results in smaller pressure deviation, but slower load response. The two different coordinated control modes are illustrated in Fig. 1. When a coal-fired power generating unit is scheduled automatically through AGC by the regional grid load dispatch center, the power plant often puts the priority in meeting the power grid

Fig. 1. Schematics of two typical coordinated control modes. (a) Boiler-following based coordinated control mode. (b) Turbine-following based coordinated control mode.

load demand to avoid additional penalty. Thus the BFCC mode, with its fast load response, is the preferred coordinated control mode under AGC control, and it is adopted by most SC/USC power units. For the 600 MW SC power unit investigated in this paper, BFCC mode is also employed. The actual control logic is shown in Fig. 2. In Fig. 2, the three cascaded f(t) functions represent three firstorder inertia links; f1(x), f2(x) and f3(x) are piecewise linearization functions of different nonlinear links; APID is a general PID controller. The turbine-side PID controller is mainly used for regulating the unit’s actual power. Meanwhile, the boiler-side PID

Fig. 2. Actual coordinated control logic of the given SC unit.

Please cite this article as: Ma, L., et al. Intelligent coordinated controller design for a 600 MW supercritical boiler unit.... Control Engineering Practice (2015), http://dx.doi.org/10.1016/j.conengprac.2015.09.002i

L. Ma et al. / Control Engineering Practice ∎ (∎∎∎∎) ∎∎∎–∎∎∎

controller is mainly responsible for maintaining the main steam pressure. In addition, in order to balance the two control variables and prevent either too large load deviation or too large pressure deviation, a pressure feedforward signal and a load feedforward signal are added to the controllers of turbine and boiler, respectively.

y*

Neura al Network Invers se System

3

u

Fig. 4. Basic structure neural network inverse system.

Distu urbances d1,…,d m

2.2. Simplified model structure of SC boiler unit When a SC coal-fired boiler unit works in the "once-through" stage, there is no clear demarcation point between steam and water. The feedwater pumped into the boiler is continuously heated, evaporated and overheated in the waterwall and all stages of superheaters. The length of each stage changes with disturbances of fuel flow rate B (kg/s), feed-water flow W (kg/s) and turbine governor valve opening μ (%), leading to changes of superheated steam temperature Ts (°C), main steam pressure Ps (MPa) and unit power Ne (MW). Thus generally a SC/USC power unit's performance can be described as a model with three inputs and three outputs, revealing the strongly coupled nonlinear relationship between the fuel flow, feed-water flow and turbine governor valve opening, and the unit's load, main steam pressure and the intermediate point temperature (heat enthalpy) (Yan et al., 2012; Liu, Kong, Hou, & Wang, 2013). Considering that the modeling purpose in this work is mainly for coordinated control to improve the control quality of unit load and main steam pressure, the intermediate point temperature (heat enthalpy) can be omitted during modeling. Thus the properties of the once-through boiler unit can be simplified into a non-linear model with three inputs and two outputs, as shown in Fig. 3.

3. Structure of load and main steam pressure inverse models 3.1. Principle of neural network inverse system model Among different nonlinear system control methods, the inverse system method is intuitive and easy to understand, which has been applied to different industrial processes. Based on neural network inverse system theory, if the inverse model u = f −1 (y ) of a typical SISO (single-input single-output) nonlinear system y = f (u) can be approximated by a neural network model and this NN inverse model is cascaded with the original system, then a quasi-linearization system y = g y* can be constructed and solved with linear system methods . As shown in Fig. 4, this kind of NN inverse model only includes the control input and the controlled variable. Therefore, it is called the basic-structure inverse system model (Dai, 2005; Ma et al., 2010). To enhance the adaptability and anti-interference performance of the inverse system model, and to widen its operation range, some important process variables and disturbances in the original system can be added to the inputs of the basic-structure neural network inverse system model. In this way, an expanded-structure

( )

Turbine Goverrnor Valve Openinng Feedwater Fllow

Fuel Flow

μ

Ne

Unit Load

W B

y

Original System

Ps Main Steam Pressure

Fig. 3. Simplified model of a SC boiler unit.

… Other Observable O Va ariables

… y*

ural Network Neu Inve erse System

u

Origin nal System

y

Fig. 5. Expanded-structure inverse system model.

neural network inverse system model is constructed, as shown in Fig. 5. After the expanded-structure neural network inverse system model shown in Fig. 5 has been established, trained and validated with high precision, a neural network inverse controller can be designed to realize inverse control of the original system.

3.2. Structure determination of the two inverse models As shown in Fig. 3, a SC boiler unit can be generally simplified as a three-input two-output model. Considering the strong coupling among the inputs and outputs and the diversity of the unit’s coordinated control modes, if this model structure is selected to build the inverse models of the load and main steam pressure, the reversibility of the model is difficult to guarantee, and it is not suitable for inverse controllers' design. In addition, the unit power is related directly to the steam temperature before the turbine governor valve, not the intermediate point temperature. The main steam temperature of a supercritical boiler unit not only depends on the feedwater-coal ratio, but also is greatly affected by the water-spray attemperators during dynamic loading process. The model in Fig. 3 does not take the water-spray attemperators into account, so it may lead to large load prediction error. For the simplified system of a SC power unit shown in Fig. 6, main steam pressure Ps (MPa), main steam temperature Ts (°C) and turbine governor value opening μ (%) are the three most important variables directly related to the turbine load. Main steam pressure Ps is affected by boiler fuel flow B, feedwater flow W and turbine governor value opening μ. To facilitate the design of the inverse coordinated controller, the inverse model structure is determined with the following considerations: (1) The main steam pressure model and the unit load model are built separately to make each model simpler in structure and reversible. (2) Since BFCC mode has faster load response under AGC than other coordinated control modes, the inverse model development is based on BFCC mode. Based on the above considerations, two separate direct models for the main steam pressure Ps and the unit load Ne are set up as shown in Fig. 7(a). Corresponding to the BFCC mode, turbine governor valve opening is responsible for adjusting the unit load, and fuel flow is for maintaining the steam pressure. Thus the two inverse models are formed as shown in Fig. 7(b). It can be easily seen that the reversibility of the two inverse models is guaranteed, and the models are with simple structure and fit for inverse coordinated controller design.

Please cite this article as: Ma, L., et al. Intelligent coordinated controller design for a 600 MW supercritical boiler unit.... Control Engineering Practice (2015), http://dx.doi.org/10.1016/j.conengprac.2015.09.002i

L. Ma et al. / Control Engineering Practice ∎ (∎∎∎∎) ∎∎∎–∎∎∎

4

Main Steam Pres. and Temp. (Ps, Ts) Reheater Steam Pres. and Temp. (Pr, Tr)

HP Turbine Governor Valve Opening μ

Reheater

Economizer

Fuel Flow B

Superheater

Boiler

Output Power Ne

Generator HP Turbine

IP Turbine

LP Turbine

Feedwater Flow W Fig. 6. Schematic diagram for supercritical power unit.

μ W

Ps Ps

Ts

B

μ

μ

Ps

W Ps

B

Ts

Ne

μ

Ne

Fig. 7. Direct/inverse models for the load and the main steam pressure of a SC unit. (μ: turbine governor value opening; W: feedwater flow; B: fuel flow; Ps: Main steam pressure; Ts: main steam temperature; Ne: output power.) (a) Direct models. (b) Inverse models.

4. Neural network inverse model development 4.1. Structure of neural network inverse models Based on the direction of information flow, artificial neural networks can be grouped into feedforward networks and recurrent networks. For a feedforward network, such as a backpropagation (BP) network or a radial basis function (RBF) network, information is one-way propagated from the input layer to the hidden layer(s) then to the output layer. This kind of neural network is also called static network. A recurrent neural network, such as an Elman network or diagonal recurrent neural network (DRNN), differs from a conventional feedforward network in which it includes recurrent or feedback connections on self or between different layers (Ku & Lee, 1995; Lee et al., 2007a). Although a dynamic network is more suitable for dynamic process modeling considering from the network structure, a static network, such as a BP network, with simpler structure, more effective learning algorithm and easy convergence performance, can also be applied for dynamic system modeling by introducing time-delayed data series of inputs and outputs into the input layer (Gencay & Liu, 1997). The BP neural network is the most successfully used ANN structure, which has been widely used in modeling and control of complex industrial dynamic systems. It is a typical feedforward neural network consisting of an input layer, one or more hidden layer(s) and an output layer. The inputs are passed forward layer by layer through weight-connected activation functions to produce outputs. Each layer contains many individual neurons and there exists no feedback between different layers of neurons.

Back-propagation (BP) algorithm is the basic algorithm to train a neural network model. During the training stage, a BP network achieves minimum mean squared error (MSE) by adjusting the weights and thresholds of the network continuously according to the error back-propagation (BP) algorithm. Because a standard BP algorithm has shortcomings such as slow convergence efficiency and easiness to fall into local minimum during model training, various improved algorithms are put forward to get fast training, such as variable learning rate algorithms, conjugate gradient algorithm, quasi-Newton algorithms and Levenberg–Marquardt algorithm, etc. In general, on function approximation problems, for networks that contain up to a few hundred weights, the Levenberg–Marquardt algorithm (trainlm) will have the fastest convergence. In many cases, trainlm is able to obtain lower mean square errors than any of the other algorithms tested (Demuth, Beale, & Hagan, 1997). Therefore, a standard BP (or feedforward) neural network with time-delayed inputs and output feedbacks is applied to establish the inverse dynamic system models for the load and the main steam pressure of a 600 MW SC boiler unit in this work. The hidden-layer and the output-layer activation functions of the BP neural network adopt Matlab purelin function and tansig function, respectively; and the Levenberg–Marquardt (LM) algorithm (trainlm) is used for model training. For the two BP network inverse models, the current and delayed values of the 3 inputs are used as the model inputs. The delayed value of the fuel flow (or turbine valve opening) is also introduced as the model’s input. The current value of fuel (or turbine valve opening) is used as the model output. Thus the two dynamic inverse models both have 7 inputs and 1 output, as shown in Fig. 8. 4.2. Training of neural network inverse models The neural network models should be trained with sample data set after the models' inputs, outputs and structures are determined. To make a NN model fully represent the dynamic and static characteristics of the controlled object, the training data should be rich enough to contain different operating conditions under which the model will be applied. In this work, a commercial-grade full-scope simulator of the given 600 MW SC boiler unit is used for modeling and control simulation tests, which is developed by a famous domestic professional power station simulation product developer, Baoding Sinosimu Technology Co. Ltd. It can simulate the whole production process of the SC power unit, including normal startup and shutdown, transient process of loading up and down, switch among different control modes, characteristics of tripping under various malfunctions, etc. All models comply with the fundamental laws of mass and energy balance, physics, math and electric power principles rather than the presetting relative curves. The model

Please cite this article as: Ma, L., et al. Intelligent coordinated controller design for a 600 MW supercritical boiler unit.... Control Engineering Practice (2015), http://dx.doi.org/10.1016/j.conengprac.2015.09.002i

L. Ma et al. / Control Engineering Practice ∎ (∎∎∎∎) ∎∎∎–∎∎∎

5

4.3. Model verification for load-changing conditions Ps (k) z-1

Input Layer (7 Neurons)

z-1

Hidden Layer (21 Neurons)

z-1

Output Layer (1 Neuron)

Ts (k)

Ne(k)

μ (k)

z-1

μ(k-1)

To verify the on-line dynamic prediction performance of the trained models, tests are carried out under a wide range of loading conditions. During loading-down process from 600 MW to 420 MW with the ramping rate of 10 MW/min, the models' realtime outputs are compared with those of the power plant simulator in Fig. 10. It can be seen that the two dynamic inverse models predict the actual outputs with very high accuracy.

5. Inverse control scheme design 5.1. Inverse compensation control scheme

μ(k) z-1

Input Layer (7 Neurons)

z-1

Hidden Layer (28 Neurons)

z-1

Output Layer (1 Neuron)

W(k) Ps (k)

B(k)

z-1

B(k-1) Fig. 8. Structure of the 2 inverse models: (a) Turbine valve opening vs. unit load. (b) Fuel flow vs. main steam pressure. (a) Load inverse mode. (b) Main steam pressure inverse model.

parameters are modulated and debugged based on the detailed specifications provided by manufactures and the real power plant on-site measurement data. Therefore, the simulator is with higher static precision and can reflect the dynamic load-changing process very well. It provides a high-fidelity experiment platform for boiler-turbine unit performance analysis, advanced control scheme design and tests before possible field application. In our work, the training data used for model training are obtained from the simulator. All of the subsequent control simulation tests are also finished with this simulator. During data acquisition, the simulator is running in BFBC mode, and the feed-water control, water-spray attemperator controls, air flow control, et al., are all put into automatic modes. 11,927 sets of data are collected from the simulator with the sampling period of 2 s, including different steady-state data between 600 MW and 420 MW load levels and the dynamic transient data between different load levels with the load ramping rate of 12 MW/min. Matlab Neural Network Toolbox functions are then used to construct the NN inverse models. Number of hidden nodes in a neural network has great influence on its performance. In this work, the hidden-layer nodes of the two neural network models are determined by trial and error. The final model structures and the mean squared errors (MSEs) of the two models after 1000 training cycles are shown in Table 1. The calculation results of the two dynamic BP network models are shown in Fig. 9. It is evident that the dynamic BP network models with time-delay inputs and output feedback are with high fitting precision. Table 1 Training results of the 2 inverse models.

After the two inverse models have been trained with sufficient accuracy, the NN inverse coordinated controllers can be constructed, which can directly replace the original PID coordinated controllers to adjust the fuel flow and turbine valve opening to keep the main steam pressure and follow the load demand. For the control action to take place, the last input of each inverse model, i.e., Ps and Ne in Fig. 6(b), needs to be replaced with the desired pressure value and the reference output of the unit load, respectively. The direct inverse control scheme is shown in Fig. 11. For a complex power generating unit, out of operation safety and reliability consideration, it is often not allowed to abandon the original control logic. Because of this, the best compromising solution is to provide a supplementary signal to the original control demand to improve the coordinated control effect (Ma & Lee, 2011). In this work, the inverse compensation control scheme is adopted by adding two supplementary signals coming from the NN inverse controllers to the original coordinated controllers’ outputs, as shown in Fig. 12. 5.2. Real-time updating of the load and pressure references A neural network inverse model itself is a kind of approximation of the inverse system. The imperfection of the model structure and the incomplete training samples will both lead to modeling error. When the NN inverse models are used as real-time controllers, the actual operating condition may also be different from the model training or validating conditions, thus producing control error. In addition, the load and main steam pressure cannot be changed instantaneously when the differences between the setpoints and current values are big. Therefore, the adoption of fixed load and pressure setpoints in the NN controllers not necessarily brings good control effect. As a solution, real-time load Ne (k) and pressure Ps(k) at the k-th step are introduced to adjust the input reference values of the NN controllers at k þ1 step automatically. The reference load and pressure values, Lref (kþ 1) and Pref (kþ 1), are adjusted by:

L ref (k + 1) = L ref (k )−k1 (Ne (k )−L sp )

(1)

Pref (k + 1) = Pref (k )−k2 (Ps (k )−Psp )

(2)

where, k1 and k2 are two saturation factors related to load error and pressure error, respectively.

6. Inverse control simulation tests

Inverse model category

Model structure

MSE

Main steam pressure Load

7-21-1 7-28-1

1.3726  10  7 3.3205  10  7

Based on the NN inverse control compensation scheme, detailed control simulation experiments are made with the fullscope simulator of a 600 MW SC boiler unit.

Please cite this article as: Ma, L., et al. Intelligent coordinated controller design for a 600 MW supercritical boiler unit.... Control Engineering Practice (2015), http://dx.doi.org/10.1016/j.conengprac.2015.09.002i

L. Ma et al. / Control Engineering Practice ∎ (∎∎∎∎) ∎∎∎–∎∎∎

6

95

240

220

200

Fuel flow (t/h)

Turbine valve opening (%)

90

85

80

75

Actual Output

180

160

140

Actual Output

Model Output

Model Output 70

0

2000

4000

6000

8000

10000

120

12000

0

2000

4000

6000

8000

10000

12000

Sample

Sample

230

94

220

92

210

90

200

88

190 180 170 160 Actual Output 150

Model Output

140

Turbine vlave opening (%)

Fuel flow (t/h)

Fig. 9. Training results of the two NN inverse models: (a) Fuel flow. (b) Turbine valve opening. (Actual output in black solid, model output in red dots.). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

86 84 82 80 Actual Output 78

Model Output

76 0

2000

4000

6000

8000

10000 12000 14000 16000 18000

Time (s)

0

2000

4000

6000

8000

10000 12000 14000 16000 18000

Time (s)

Fig. 10. On-line test of the NN inverse models: (a) Fuel flow, (b) Turbine valve opening. (Actual output in black solid, model output in red dots.). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Measured Variables and Disturbances

…

yref(k+1) NN Inverse ur(k+1) Controlled y(k+1) Object Controller Load Load Reference Turbine Valve Demand (Pressure Reference) (Coal Feeder Demand) (Pressure) Fig. 11. Direct inverse coordinated control scheme.

When the unit load is changed in turn from 600 MW, to 540 MW, to 480 MW, and then back to 540 MW and to 600 MW, with the load ramping rate of 12 MW/min and the pressure changing rate of 1 MPa/min, the NN inverse compensation control results are compared with the original coordinated control results in Fig. 13. It can be seen from Fig. 13 that, with the original coordinated control scheme, the load and pressure control effect is apparently different at different load level. The control quality deteriorates gradually with load decreasing. The maximum load deviation is over 7 5 MW and the maximum pressure deviation is over 71.5 MPa. The control overshoot is relatively large and the stabilizing time is long. With the NN inverse compensation control scheme, during the whole loading scope, the actual unit load can track the load command very well. The maximum pressure

Fig. 12. Inverse compensation control schematic.

deviation is less than 71.0 MPa, and the stabilizing time is shorter. For the above experiments, the load changing rate and the pressure changing rate are consistent with the model training condition. To test the inverse coordinated control effect under an operating condition different from the training samples, the load ramping rate is changed to 6 MW/min, and the pressure-changing

Please cite this article as: Ma, L., et al. Intelligent coordinated controller design for a 600 MW supercritical boiler unit.... Control Engineering Practice (2015), http://dx.doi.org/10.1016/j.conengprac.2015.09.002i

620

620

600

600

580

580

560

560

Load (MW)

Loa d (MW )

L. Ma et al. / Control Engineering Practice ∎ (∎∎∎∎) ∎∎∎–∎∎∎

540 520 500 Load Command Original PID Control NN Inverse Control

520

0

2000

4000 6000 Time (s)

460

8000

26

26

25

25

24 23 22 21 20

PressureSet Value OriginalPIDControl NN Inverse Control

19 0

2000

4000 6000 Time (s)

8000

Fig. 13. Control tests under verification condition: (a) Load. (b) Main steam pressure. (Reference in blue solid, PID control in green dash, NN control in red dots.). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

rate is kept at 1 MPa/min. Again, load is dropped from 600 MW to 480 MW then back to 600 MW. The load and pressure control results are compared in Fig. 14. It can be seen from Fig. 14 that, under the validation condition, the NN inverse control also achieves better results than the original PID control. The load can completely track the load instruction and the pressure deviation is less than 71.0 MPa. The overshoot is smaller and the stabilizing time is shorter.

Load Command Original PID Control NN Inverse Control

480

Main steam pressure (MPa)

Main steam pressure (MPa)

540

500

480 460

7

0

2000

4000

6000 8000 10000 12000 Time(s)

24 23 22 21 Pressure Set Value Original PID Control NN Inverse Control

20 19 0

2000

4000

6000 8000 Time(s)

10000 12000

Fig. 14. Control tests under validation condition: (a) Load. (b) Main steam pressure. (Reference in blue solid, PID control in green dash, NN control in red dots.). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

It should be pointed out that a supercritical power unit is a complicated nonlinear system. There are many factors influencing the load and main steam pressure. Selection of input variables, neural network structure, and the choice of training data, all have influence on the final control effect. In addition, potential difficulties when applying the approach to a real power plant rather than a simulator lie in that, more possible disturbances may exist, such as changing air–fuel ratio, varying fuel composition or different auxiliary systems. Under these cases, real-time retuning (or at least regular tuning) of the NN inverse models becomes quite necessary, which still needs further research.

7. Conclusion To improve the coordinated control quality of a supercritical power generating unit, the dynamic feedforward neural network inverse models for the load and main steam pressure (MSP) are developed. Based on the trained models, the NN inverse coordinated controllers are designed, programmed and tested in the full-scope simulator of a 600 MW SC power unit. It is shown that the proposed NN inverse coordinated control method has better performance in load responsiveness, steam pressure overshoot and control accuracy, compared to the original PID scheme.

Acknowledgements This project is supported by the National Natural Science Foundation of China under Grant 61174111. References Dai, X., He, D., Zhang, T., & Zhang, K. (2003). ANN generalized inversion for the linearization and decoupling control of nonlinear systems. IEE

Please cite this article as: Ma, L., et al. Intelligent coordinated controller design for a 600 MW supercritical boiler unit.... Control Engineering Practice (2015), http://dx.doi.org/10.1016/j.conengprac.2015.09.002i

8

L. Ma et al. / Control Engineering Practice ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Proceedings – Control Theory and Applications, 150(3), 267–277. Dai, X. (2005). Multivariable nonlinear neural network inverse system control method. Beijing: Science Press. Demuth, H., Beale, M., & Hagan, M. (1997). Neural network toolboxTM 6 'guide. USA: The Math Works, Inc. Ding, J., Li, G. Y., & Wang, N. (2011). Research on simplified model for the control of 660 MW supercritical unit. Electric Power Science and Engineering, 27(5), 29–34. Garduno, R., & Lee, K. Y. (2005). Compensation of control-loop interaction for power plant wide-range operation. Control Engineering Practice, 13, 1475–1487. Garduno, R., & Lee, K.Y. (2000). Feedforward compensated multiloop control of a power plant. In Proceedings of the 2000 IEEE power engineering society summer meeting. Seattle, USA. Garduno, R., & Lee, K.Y. (2002). Power plant coordinated control with wide-range control-loop interaction compensation. In Proceedings of the IFAC 15th world congress. Barcelona, Spain. Gencay, R., & Liu, T. (1997). Nonlinear modeling and prediction with feedforward and recurrent networks. Physica D, 108, 119–134. Gerhard, K. L. (1998). Improved maneuverability of power plants for better grid stability. Control Engineering Practice, 6(12), 1549–1557. Han, P., Zhang, H. L., & Zhang, L. J. (2001). Simulation study on neural network based adaptive inverse control. Journal of North China Electric Power University, 28(3), 26–30. Heo, J. S., & Lee, K. Y. (2008). A multi-agent system-based reference governor for multiobjective power plant operation. IEEE Transactions on Energy Conversion, 23(4), 1082–1092. Jan, H. M., Tommy, M., Palle, A., & Tom, S. P. (1998). Optimization of boiler control to improve the load-following capability of power-plant units. Control Engineering Practice, 6(12), 1531–1539. Jia, D. X., Cheng, H. Z., Xiong, H. G., et al. (2007). Financial compensation of AGC unit considering control performance standard. Proceedings of the CSEE, 27(31), 62–66. Ku, C. C., & Lee, K. Y. (1995). Diagonal recurrent neural network for dynamic systems control. IEEE Transactions on Neural Networks, 6, 144–156. Lee, K.Y., Heo, J.S., Hoffman, J.A., & Kim, S.H. (2007a). Neural network-based modeling for a large-scale power plant. In Proceedings of the IEEE power energy society general meeting (pp. 1–8). USA. Lee, K.Y., Helo, J.S., Hoffman, J.A., & Kim, S.H. (2007b). Modified predictive optimal control using neural network-based combined model for large-scale power plants. In Proceedings of the IEEE PES general meeting (pp. 1–8). Lee, K.Y., Ma, L.Y., & Boo, C.J. (2009). Inverse dynamic neuron-controller for superheater steam temperature control of a large-scale ultra-supercritical (USC) boiler unit. IFAC Symposium on Power Plants and Power Systems Control. Tampere, Finland. Lee, K. Y., Van Sickel, J. H., Hoffman, J. A., Jung, W. H., & Kim, S. H. (2010). Controller design for a large-scale ultrasupercritical once-through boiler power plant. IEEE Transactions on Energy Conversion, 25(4), 1063–1070. Li, D., Liang, J., Sun, R. F., Liu, J., Liu, G. Q., Chen, Z. X., & Han, F. K. (2010). Compensation strategies of AGC regulation performance in plants management and

assessment system. Automation of Electric Power Systems, 34(4), 107–111. Li, W. H., & Wang, Y. S. (2005). Analysis on control characteristics of high capacity thermal power generating sets under AGC mode. Power System Technology, 29 (18), 12–18. Liu, X. J., Kong, X. B., Hou, G. L., & Wang, J. H. (2013). Modeling of a 1000 MW power plant ultra super-critical boiler system using fuzzy-neural network methods. Energy Conversion and Management(65), 518–527. Ma, L. Y., Shi, Z. X., & Lee, K. Y. (2010). Superheater steam temperature control based on the expanded-structure neural network inverse models. Advanced Material Research, 433–444, 401–407. Ma, L.Y., & Lee, K.Y. (2011). Neural network based superheater steam temperature control for a large-scale supercritical boiler unit. In Proceedings of the IEEE PES General Meeting. Detroit, USA. Ma, L.Y., Lee, K.Y., & Ge, Y.P. (2012). An improved predictive optimal controller with elastic search space for steam temperature control of large-scale supercritical power unit. In Proceeding of 51st IEEE conference on decision and control (pp. 7024–7029). December 10–13, Maui, Hawaii, USA. Ma, L.Y., Wang, Z.Y., & Lee K.Y. (2014). Neural network inverse control for the coordinated system of a 600 MW supercritical boiler unit. IFAC2014. Malinowski, A., Zurada, J.M., & Lilly, J.H. (1995). Inverse control of nonlinear systems using neural network observer and inverse mapping approach. In Proceedings of IEEE International Conference on Neural Networks (pp. 2513–2518). Vol. 5. Perth, Western Australia. Wang, D. F., Yu, X. N., & Song, Z. P. (2002). Dynamic mathematical model of ball mill pulverizing system and its inverse control based on distributed neural networks. Proceedings of the CSEE, 22(1), 97–101. Wang, L. J., & Li, X. M. (2011). The optimization of coordinated control system under AGC mode for supercritical generation unit based on “two documents”. Modern Electric Power, 28(2), 84–89. Widrow, B., McCool, J., & Medoff B. (1978). Adaptive control by inverse modelling. In Proceedings of the 12th Asimolar conference on Circuits, Systems and Computers. Widrow, B. (1986). Adaptive inverse control. In Proceedings of the Second IFAC Workshop on Adaptive Systems in Control and Signal Processing (pp. 1–5). Lund, Sweden: Lund Institute of technology. Widrow, B., & Walach, E. (1996). Adaptive inverse control. Upper Saddle River, NJ: Prentice Hall PTR. Xie, S.Y., Zhang, C.J., & Xiao, X.L. (2007). Adaptive inverse induction machine control based on variable learning rate BP algorithm. In Proceedings of the IEEE international conference on automation and logistics. Jinan, August 18–21. Yan, S., Zeng, D. L., Liu, J. Z., & Liang, Q. J. (2012). A simplified non-linear model of a once-through boiler-turbine unit and its application. Proceedings of the CSEE, 32 (11), 126–134. Yuan, Z.R., & Guo, X.G. (1994). Back-propagation neural networks for the inverse control of discrete-time nonlinear plant. In Proceedings of the American Control Conference. Baltimore, Maryland, June 1994. Zhang, Z. Y., Li, W. H., & Song, Z. X. (2007). Control strategy of coordination control system of 600 MW unit with supercritical monotube boiler. North China Electric Power, 1, 24–28.

Please cite this article as: Ma, L., et al. Intelligent coordinated controller design for a 600 MW supercritical boiler unit.... Control Engineering Practice (2015), http://dx.doi.org/10.1016/j.conengprac.2015.09.002i