IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 13, NO.

6, NOVEMBER 2005

943

A New Coordinated Control Strategy for Boiler-Turbine System of Coal-Fired Power Plant
Shaoyuan Li, Senior Member, IEEE, Hongbo Liu, Wen-Jian Cai, Yeng-Chai Soh, and Li-Hua Xie
AbstractThis paper presents the new development of the boiler-turbine coordinated control strategy using fuzzy reasoning and autotuning techniques. The boiler-turbine system is a very complex process that is a multivariable, nonlinear, slowly time-varying plant with large settling time and a lot of uncertainties. As there exist strong couplings between the main steam pressure control loop and the power output control loop in the boiler-turbine unit with large time-delay and uncertainties, automatic coordinated control of the two loops is a very challenging problem. This paper presents a new coordinated control strategy (CCS) which is organized into two levels: a basic control level and a high supervision level. Proportional-integral derivative (PID) type controllers are used in the basic level to perform basic control functions while the decoupling between two control loops can be realized in the high level. A special subclass of fuzzy inference systems, called the Gaussian partition with evenly (GPE) spaced midpoints systems, is used to self-tune the main steam pressure PID controllers parameters online based on the error signal and its rst difference, aimed at overcoming the uncertainties due to changing fuel caloric value, machine wear, contamination of the boiler heating surfaces and plant modeling errors. For the large variation of operating condition, a supervisory control level has been developed by autotuning technique. The developed CCS has been implemented in a power plant in China, and satisfactory industrial operation results demonstrate that the proposed control strategy has enhanced the adaptability and robustness of the process. Indeed, better control performance and economic benet have been achieved. Index TermsBoiler-turbine coordinated control strategy, decoupling control, industrial application, multivariable systems, power plant.

Differential parameter in PID controller. Transfer function output to input . Decoupling compensator output matrix. Decoupling compensator matrix. Error between set-point and current value. Change value of the . First-order lag lter time constant. System sampling period. Fuzzy membership of variable . System static gain. System critical period. Amplitude margin. Phase margin. Boiler ring rate. Governor value position. Main steam pressure. Power output. Main steam ow. Main steam temperature. Abbreviations DCS Distributed control systems. CCS Coordinated control system. GPE Gaussian partition with evenly space. TPE Triangle partition with evenly space. CARMA Controlled autoregressive moving average. AGC Automatic generation control. II. INTRODUCTION

I. NOMENCLATURE Critical gain in power control loop. Critical gain in pressure control loop. Proportional parameter in proportional-integral derivative (PID) controller. Integral parameter in PID controller.
Manuscript received December 1, 2003. Manuscript received in nal form June 8, 2005. Recommended by Associate Editor V. Gopal. This work was supported in part by the National Natural Science Foundation of China under Grant 60474051, in part by the Key Technology and Development Program of Shanghai Science and Technology Department under Grant 04DZ11008, and in part by the program for New Century Excellent Talents in University of China (NCET). S. Li and H. Liu are with the Institute of Automation, Shanghai Jiaotong University, Shanghai 200030, China (e-mail: syli@ sjtu.edu.cn). W.-J. Cai, Y.-C. Soh, and L.-H. Xie are with the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798, Singapore (e-mail: ewjcai@ntu.edu.sg). Digital Object Identier 10.1109/TCST.2005.854319

HE majority of coal-red power plants in China built more than a decade ago were once expected to operate at near full capacity but are now operating in a load following mode due to the rapid development of the power industry. Consequently, AGC of power networks becomes necessary in order to meet varying load demands at different time periods. As a key component of power network AGC, the coordinated control of fossil-fueled generating units plays a vital role in safe and economic operation of the system. Since the performance of the multi-inputmulti-output (MIMO) boiler-turbine system can vary signicantly due to the complex nonlinearity in different operating regions, conventional linear control methods may not be sufcient for the whole operation range [1][5]. For the boiler-turbine control system, the central task is to adjust the output power to meet system demand while minimizing unwanted pressure and temperature variations. The turbine speed is controlled by the main steam pressure to drive

1063-6536/$20.00 2005 IEEE

944

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 13, NO. 6, NOVEMBER 2005

the power generator [1]. However, as the electricity demand swings from minute to minute, the turbine speed has to be varied to meet the different load demands. The variation of turbine speed causes a chain reaction of change in ring rate, demand on the coal, grinding and feeding coal, and eventually the steam generated. The time scale of the steam pressure process is naturally quite slow at about 8 to 15 min for the system under study. In contrast, by opening the governor valve, different amount of steam can be supplied immediately [2], but this is at the expense of depleting stored energy in the evaporator of the boiler leading to main steam pressure variations, and it takes a long time to recover it to its reference value. Thus, for a rapidly changing power demand, controlling the governor valve will be more effective, but a sustained change can only be achieved via changing the ring rate. As a compromise, a masterslave control strategy [3] has been adopted in most power plants in conventional control mode, i.e., the turbine speed control loop works as the master loop to track the main steam pressure, and the combustion control loop as the slave loop to track the varying turbine speed. In such a scheme, a proportional derivative (PD) controller is needed to coordinate the relationship between the two control loops [3], the controller parameters, however, are very difcult to adjust even within a very narrow operation range. In practice, it is mostly done by a trial-and-error method. As a result, the main steam pressure and the power output of the generating units, in most systems, are still being controlled manually. It is difcult to avoid excessive stresses on the process components and to meet the economic and quality requirement. Even though several advanced control structures for the boiler-turbine CCS have been proposed to tackle the problem in the literature, the problem still remains unsolved for a large operating range [6][9]. In this paper, an advanced control strategy is proposed to solve this particular coordinated control problem. By considering the complex nature of the process, a supervisory control structure is proposed and it consists of the following components. 1) Proportionalintegral derivative (PID)-type controllers are used as basic control units due to their simple structure and the concept is well understood by eld engineers and operators. 2) A steady-state triangular decoupler is designed which simultaneously decouples the strongly coupled main steam pressure and power output loops for both set-point and unmeasured pulverized coal disturbance. 3) A supervisory control level via an autotuning technique is used to tune the parameters of the controller at different operating conditions as system parameters can vary signicantly due to the complex nonlinearity in load following mode. 4) The PID type controllers and decouplers are gain-scheduled according to the actual load to take account of the load-dependent nonlinear characteristics of the boiler-turbine process. 5) A special subclass of fuzzy inference systems, i.e., Gaussian partition system with evenly spaced midpoints (GPE) [17], is employed to autotune the PID parameters

Fig. 1. Schematic diagram of the 300 MW boiler-turbine unit.

Fig. 2. Conventional CCS.

of the main steam pressure online, and it overcomes uncertainties caused by changing fuel caloric value, machine wear and plant modeling errors. The control strategy has been realized in the FOXBORO I/A Series DCS and implemented on a 300 MW boiler-turbine unit, i.e., Unit 1 of Yuanbaoshan Power Plant in China for two years. The remaining sections of this paper are organized as follows. Section II describes the process and plant characteristics. The steady-state triangular decoupler and autotuning supervisory control techniques are given in Section III. Section IV discusses the gain-scheduling of the PID type controllers and decouplers and the autotuning main steam pressure PID controller by a fuzzy mechanism. Section V presents the implementation and the operating results of the proposed control strategy on the boiler-turbine system. Finally, conclusions are drawn in Section VI. III. PROCESS DESCRIPTION The schematic diagram of the investigated 300 MW boilerturbine unit is shown in Fig. 1. The boiler is a compound circulation tower boiler with low circulation ratio that produces 947 tons of steam per hour at maximum continuous rating. The rated main steam pressure and the super-heater outlet temperature are 18.5 Mpa and 545 C, respectively. Electric power is generated by the feeding steam from the boiler to the turbine, and the steam from the turbine condenses to water through the condenser, which is sent to the boiler again by the feedwater pump.

LI et al.: A NEW COORDINATED CONTROL STRATEGY FOR BOILER-TURBINE SYSTEM

945

Fig. 3.

New CCS.

A characteristic of the boiler system is highly complex and nonlinear. The main nonlinearity is related to the property of gain and time constant in the boiler process, and the property variation is dependent on the plant load. In order to control the complex system, the boiler-turbine CCS is usually adopted in the industry which can be regarded as a two-inputtwo-output (TITO) multivariable control system. The conventional structure of CCS is shown in Fig. 2, where the two inputs are boiler ring rate (coal feeder speed) and turbine governor valve poand the sition ; two outputs are the main steam pressure , respectively. real power output The boiler-turbine unit is a time-varying and nonlinear system with strong interactions and uncertainties. Through careful theoretical analysis [5], a linearized system model for a given operating point can be obtained as

(1) denotes the steady-state values of and where the and , is the transfer function between the main steam ow and the power output which represents the dynamics of the turbine and reheater. Although the boiler-turbine unit transfer function can be approximated by (1) for a given operating point, the boiler-turbine system remains very complex, and conventional control approaches [1], [5] have encountered great difculties due to the following factors. 1) Strong Coupling. There are strong couplings between the main steam pressure control loop and the power output control loop. If there exists frequent unmeasured

pulverized coal disturbances caused by uncertain coal mill working conditions [3], the strong couplings could create severe problems for system stability and control, particularly when the unit is operating at high load. 2) Nonlinearity. The system exhibits highly nonlinear characteristics when the power output changes over a wide range, as pressure increases/decreases and heat transfer rates do not vary in proportion to the fuel inputs. 3) Long Settling Time with Time-delay. There is a long settling time in the change of steam pressure or power output while changing the coal feeder speed due to massive storages of mass and energy. The results of real-time dynamic experiments [5] show that the time constant and time-delay from the coal feeder speed to the main steam pressure or the power output can vary within 6 15 min with varying load. 4) Uncertainty. In coal-red units, the fuel is supplied by coal mills which have uncertain dynamics due to varying time-delay of the grinding process and the uncertain behavior of the mills caused by coal quality variation and machine wear. IV. COORDINATED CONTROL A hierarchical coordinated control scheme for the main steam pressure and the power output consists of two levels: a basic control level and a supervisory control level as shown in Fig. 3. The supervisory control level is referred to as the autotuner of CCS controller/decoupler, its functions include: monitoring the control performance online and identifying the mathematical model of the boiler-turbine system;

946

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 13, NO. 6, NOVEMBER 2005

Fig. 4. Multivariable feedback structure of CCS.

retuning CCS controller/decoupler parameters automatically to enhance the adaptability of the control system due to large variation of operating conditions. The basic control level consists of three conventional feedforward/feedback PID-type controllers: A unilateral PD-type decoupler adjusts the power output to follow load demand changes while maintaining the main steam pressure within the permitted range and compensates the load varying caused by the uncertain coal mill working condition. A PI controller for the electric power output loop. A fuzzy autotuning PID controller in the main steam pressure control loop which provides good performance with the ability to overcome uncertainties due to changing fuel caloric values, machine wear and plant modeling errors. A. Steady-State Triangular Decoupler As the time-delay of the turbine and reheater is much smaller than that of the boiler , the time-delay in can be ignored and the following approximation in (1) can be obtained: (2) . Fig. 4 shows the block diwhere is the static gain of agram of the multivariable feedback control structure of CCS, is the matrix containing the PID is the input compensator matrix and is the controllers, output compensator matrix. Denote respectively, as , If and are chosen,

Fig. 5. Z-N identication scheme of CCS.

1) Closed-loop identication of the controlled plant by using the decentralized step test and least-squares algorithm around an operating condition based on the inputoutput plant data (see the Appendix). in (2) according to the con2) Determine the static gain trolled plant model to adapt the decoupling compensator matrix . 3) Perform multivariable ZieglerNichols identication experiments [16] according to Fig. 5 using the dynamic model obtained from Step 1) to determine the ultimate and of the conperiod , the ultimate gains trolled plant (see the Appendix) corresponding to the power output control loop and the main steam pressure control loop, respectively. and , determine prescribed ranges 4) Based on , ) and ( , ) for the ( main steam pressure fuzzy autotuning controller. In this application, based on the experiments, we , , choose , . , and , use the tuning 5) From the values of method proposed in [16] to tune the PI parameters of the power output control loop.

V. BASIC CONTROL LEVEL and (3) The boiler-turbine process includes fairly strong nonlinearities due to the difference between the stored energy at each plant load. To cope with the load-dependent nonlinear dynamic characteristics, in the basic control level the local PID type controllers and decouplers are designed at several load levels based on linear control theory rstly according to methods described in Section III-B, and then gain-scheduling technique according to the actual load described in Section IV-B is used to obtain the global controllers/decouplers parameters. There are three PID-type controllers in the basic control level as shown in Fig. 3, each has different functions. and obtained from Step 3) of 1) PI controller Form and can the autotuning algorithm, the parameters be obtained by ZieglerNichols method as follows [16]: (5) 2) PD controller A PD controller as shown in Fig. 3 provides a feedforward compensation to the given load

and substituting (2) into (1), the transfer function matrix of CCS for the controlled plant becomes (4) The strong interaction between the two control loops is reduced, and the process is now approximately a triangular decoupled system. The parameter of the decoupling compensator matrix can be adapted by the CCS autotuner. B. Autotuning CCS Parameters The autotuner for CCS parameters monitors the control performance of the CCS online. If the absolute value of the error between the set-point and the current value of the main steam pressure exceed a prescribed threshold for a predetermined period of time, the autotuner automatically retunes the CCS parameters. Otherwise, the CCS parameters remain unchanged. The autotuning algorithm is given as follows.

LI et al.: A NEW COORDINATED CONTROL STRATEGY FOR BOILER-TURBINE SYSTEM

947

under the Voltage-constant running mode. From Figs. 2 or 3, the equivalent transfer function of the system with in the generator is given by an integral lter (6) With a feedforward controller to compensate the action of to , we have (7) where is the controller in the rst control loop, and feedforward controller can be obtained as (8) This feedforward controller may not be realizable. In practice, a PD controller is usually adopted to satisfy this need with the following form: (9) and can be calculated by where the parameters the ZieglerNichols method in [16], as follows: (10) can be determined from and the lter time constant experiments, the engineering rule of thumb value is two to three times the process time constant. 3) PID controller The PID controller used in this paper is given as follows: (11) where the time constant of the rst-order lag lter , and , . Through onsite observation and theoretical analysis [5], the main contributing factors for uncertain system performance of the main steam pressure loop are as follows. Disturbance caused by frequent unmeasured pulverized coal disturbance under the uncertain coal mill working conditions. Parameter uncertainty and variation caused by the uncertainties due to the changing fuel caloric value, machine wear and the contamination of the boiler heating surfaces. As the main steam pressure control loop plays a crucial role in rejecting various disturbances and tracking the power output, high-quality control performance is very important since the change of main steam pressure represents the energy balance between boiler steam production and grid load demand or stored energy in the evaporator. If the main steam pressure is too high, the stress on the plant increases and the plants life span will decrease. If it is too low, the efciency will decrease. If there exist persistent large main steam pressure variations, the safe operation of the unit cannot be guaranteed. When the boiler is operated at constant pressure, the main steam pressure dynamics changes with unit load, a better control strategy

is to use autotuning PID controller to implement real-time control. A. Fuzzy Autotuning PID Controller In order to enhance the robustness and control performance of the main steam pressure loop, a fuzzy autotuning PID controller is adopted for better performance [9]. By using fuzzy rules based on expert knowledge to adjust PID parameters which are initially determined by classical tuning rules, high-quality control performance can be expected than that of the PID controllers with xed parameters. 1) Fuzzy Rules for Tuning PID Parameters: Based on the step response analysis, human expertise for the process and extensive simulation studies, a set of autotuning rules for the PID parameters of the following form are proposed: if is and then is is is is (12)

where and are the current error and its rst differand represent ence of the main steam pressure, , , , , , , and , respeca member of fuzzy sets for . tively, For the given prescribed minimum/maximum of and of in Step 4) of the , and of autotuning algorithm, the tuning coefcients , and , are given, respectively, by (13) (14) (15) According to the ZieglerNichols PID tuning rule, when 4, the integral action of PID controller is moderate. Therefore, by chosen smaller or bigger than 4, a stronger or weaker integral action can be obtained. and may be either big (B) or small (S) The fuzzy sets and are characterized by the membership functions of natural logarithm. The grade of the membership functions and the or ) has the following relation: variable ( or or for big for small (16) (17)

Based on the processes characteristics, four singleton membership functions denoting the linguistic variable small (S), middle small (MS), middle (M) and big (B) for the fuzzy sets are de2.8, 3.4, 4 and signed. They are dened as 5. For the system performance to be robust and to achieve fast response with small overshoot, the PID controller must be tuned such that the three control actions coordinate with each other as the process is affected by multiple time-delays and large settling times. For example, to obtain a desired step response, a big control signal is needed at the beginning to achieve a fast rise time, which requires a big proportional gain and a small derivative

948

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 13, NO. 6, NOVEMBER 2005

TABLE I FUZZY AUTO TUNING RULES OF K

TABLE II FUZZY AUTO TUNING RULES OF K

gain. Also, since processes have large time-delay, a small integral gain is desired to reduce the overshoot. When the output response is near the set-point, the proportional gain and integral gain should be changed from large to small and from small to large, respectively, to make the controlled output converge to , and the set-point quickly. The autotuning rules for are given in Tables IIII, respectively. 2) Fuzzication Strategy and Fuzzy Inference: In fuzzication, the shape of membership function characterizes the intuition of converting the crisp value into linguistic value to t the human thinking process. Gaussian-shaped membership function is chosen as the membership function of the antecedent part or in (12) as it better ts human intuition. Similar to the triangle partition system with evenly space (TPE) [18], Gaussian-shaped membership function have the following features: 1) the membership function is symmetrical about its central value, 2) all membership functions have the same shape, and 3) the space between the central values of two adjacent members are equal and these constitute the GPE used in our system. The Gaussian membership function is dened as (18) others is the center value of the membership function of . Point is a unique element that has membership value 1 in , this guarantees four rules without zero contribution at any one time. given in (12), According to the fuzzy autotuning rule or and Tables IIII, the crisp sampling value is rst fuzzied into linguistic value based on the previous fuzzication strategy by seven reference membership functions denoting the linguistic variables: negative big (NB), negative middle (NM), negative small (NS), zero (Z), positive small (PS), positive middle (PM) and positive big (PB), respectively. given in (12) is Then, the crisp value of the th rule obtained by the product of the membership function values of and (19) where is the membership function value of the refergiven a value of , and is the ence fuzzy set given membership function value of the reference fuzzy set . Based on , the values of and for a value of each rule are determined from their corresponding membership where

TABLE III FUZZY AUTO TUNING RULES OF 

functions given in (8) and (9). Then the defuzzication yields the following results:

(20)

(21)

(22)

or is the value of where grade for the th rule . Once , the PID controller parameters online based on (13)(15).

or corresponding to the , and are obtained, and can be autotuned

B. Gain-Scheduling of the PID Type Controllers and Decouplers If it is known how the dynamics of a process change with the operating conditions of the process, it is possible to change the controller parameters accordingly, known as gain-scheduling. A measurable process variable, which is descriptive of the operating condition and used to adjust the controller parameters, is known as a scheduling variable. For the boiler-turbine process, is chosen as the scheduling variable. the actual load , containing m values of A set the scheduling variable is chosen and arranged according to: for ). For each value of

LI et al.: A NEW COORDINATED CONTROL STRATEGY FOR BOILER-TURBINE SYSTEM

949

in the set a linear model ( , 2) is identied by using identication methods described in Section III-B and for each model parameters of the local PID type controllers and decouplers are tuned according to methods described in Section III-B. By assuming that the parameters of the PID type controllers and decouplers change linearly between the two load levels, the parameters can be gain-scheduled between the frozen operating points by linear interpolation

TABLE IV MODEL PARAMETERS OF 210 MW

(23) according to the actual load , where .

TABLE V MODEL PARAMETERS OF 270 MW

VI. INDUSTRIAL IMPLEMENTATION AND APPLICATION RESULTS The proposed coordinated control strategy described in this paper is realized by using the software development system of the FOXBORO I/A Series DCS and has been developed and successfully applied in the control of a 300 MW boiler-turbine unit, i.e., Unit 1 of Yuanbaoshan Power Plant in China. A. Identifying the Plant Models and Autotuning Controller Parameters To design an efcient control system for such a complicated plant reliably, open-loop identication of the control plant was performed under different coal quality and load levels by using identication methods described in Section III-B. For example, parameters at two typical operating conditions of 210 MW (70% of full load) and 270 MW (90% of full load) and with caloric value of the coal in the range of 3300 to 3600 kcal/kg are obtained and listed in Tables IV and V. The form of these linear models is given in (1) where ( 1, 2). Remark 1: The identied plant models have been extensively validated, including eld dynamic experiments carried out before the control system have been designed, the open-loop identication by using the methods described in Section III-B and through comparisons between simulation results and actual plant data. From the validations, it has been concluded that the plant models reect the process dynamics in a satisfactory manner, so that the initial design of the control system can be performed on the plant models. Remark 2: The identied plant models are only accurate at the typical operating conditions. In order to avoid the disadvantages of self-tuning control based on rational transfer function parameter estimation directly, increase the robustness and facilitate commissioning, self-tuning control method from the frequency domain approach [19] is adopted by the estimation of the ultimate point of the controlled plant. Simulation studies show that the changes of the ultimate point parameters of the controlled plant are small when there are large changes in the parameters of the plant transfer function model. The results point to the robustness of the proposed self-tuning control method. To illustrate the design procedure of the proposed CCS, the local decoupler/controllers initial parameters of the two typical

operating conditions described previously were designed for instance as follows. The steady triangular decoupler parameter was tuned from the steady-state plant model parameters as 0.936 and 0.962 corresponding to the 210 MW and 270 MW operating conditions, respectively. 1 of (6) was selected. and The desired critical period and critical gains ( , ) were determined as (727 s, 0.269 and 2.304) and (556 s, 0.192 and 2.046) corresponding to the 210 MW and 270 MW operating conditions, respectively. , ) of the power output The PI controller parameters ( control loop were tuned as (1.019, 60.58 s) and (0.598, 49.86 s), respectively, for the two operating conditions ac, and based on speccording to the values of and specied amplitude ied phase margin margin 3. Based on simulation studies and real-time experiments, , ) and ( , ) the ranges for ( , , are obtained as , , respectively.

950

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 13, NO. 6, NOVEMBER 2005

Fig. 6. CCS at load change rate of 3 MW/min.

Fig. 7.

CCS at load change rate of 10 MW/min.

After the local PID type controllers/decouplers are designed at several typical load levels based on linear control models, the controller/decouplers parameters of other operating conditions are gain-scheduled according to method given in Section IV-B. B. Real-Time Application and Discussion After commissioning, the proposed CCS has been put into service in the 300 MW commercial power-generating unit for more than three years. Here, several real-time operation results are recorded and used as examples to demonstrate the effectiveness of control strategy. Case 1: An essential experiment for the boiler-turbine unit is increasing the load at the rate of 3 MW/min to test the control performance, before the CCS can be put in operation. Fig. 6 shows the operation results for a 180-min period, where the and the power output increase main steam pressure from 13.56 MPa to 18.06 MPa and from 185 MW to 285 MW, respectively. The controlled variables showed fast response with small tracking error and good steady-state performance for both the main steam pressure and the power output, while the main still maintains the set-point of 545 C. steam temperature Case 2: Load changes over a wide range between 185 MW and 285 MW at the allowable maximum rate of 10 MW/min. In this case, the main steam temperature will be severely affected under conventional control strategy. Fig. 7 shows three typical operating conditions within the specied operation range; and 1) decreasing the set-point of two control loops during 2040 min; 2) setting and as pulse signals during and during 150180 min. 90100 min; and 3) increasing and , as well as main steam The controlled variables are all kept within acceptable limits near temperature their set-points and good tracking performance is obtained. Since the maximum rate of load changes in the power plant is 6 MW/min and the system has good tracking performance even at is 10 MW/min (150160 min.), the proposed control strategy satises the load following requirement.

Fig. 8. CCS at load change rate of 9 MW/min.

Case 3: Fig. 8 shows the operation result under coordinated control when the unit is decreasing the load over a wide range at the rate of 9 MW/min from 287 MW to 197 MW. As can be seen, the principal controlled variables of the unit are all kept near their setpoints and good load following is achieved, so the efciency of the unit is enhanced. Case 4: The 4-h operation results of the proposed CCS and manual operation under high load condition are compared during commissioning to validate the proposed control scheme. As shown in Figs. 9 and 10, the main steam pressure and the power output under the proposed control strategy are running much smoother than those of manual operation under the full load 300 MW. In manual control mode, the frequent strong unmeasured pulverized coal disturbance results in the main steam pressure exceeding the prescribed upper limit for safe operation and cause the high-pressure bypass valve to open resulting in wasted energy. Under automatic control mode, the system has a very good disturbance rejection property, safe and economic operation of the unit is guaranteed. Remark 3: This paper is based on the practical application project which had been carried out at Yuanbaoshan Power Plant, China. While many of the control loops were regulated

LI et al.: A NEW COORDINATED CONTROL STRATEGY FOR BOILER-TURBINE SYSTEM

951

Fig. 11.

Closed-loop TITO control system.

Fig. 9.

CCS under manual control.

2) both the main steam pressure and the electric power output have fast responses with low overshoot and good steady-state performance; 3) the main variables of the unit including the main steam temperature are all kept within acceptable limits near their set-points and good tracking performance even under load-variation over a wide range. The CCS proposed can be easily implemented in other coalred boiler-turbine unit of the power plants without much modication. The research work on the extension of the technology and an adaptive predictive control for the CCS of power plant is currently under investigation. APPENDIX RECURSIVE LEAST SQUARES IDENTIFICATION ALGORITHM The simple identication approach is on the basis of a MIMO process under decentralized control. To simplify our derivation, we adopted two inputs and two outputs control system (shown in Fig. 11). controllers, noises and process transfer Where , and 1, 2, are functions, respectively, the notation , , and could used in both and domain. In general, be any type of controllers that make the closed-loop system stable. Without loss of generality, assumed that the controllers and are proportional type. The fundamental relationship between error signals and transfer function outputs for the system are expressed as and Assumed that the process initially test at a steady state with initial set point, error, and output variables. Notes that , , , , and , respectively, and then , . To identify the process parameters, the test involves the following two steps. 1) When kept xed, make a step change from to , record the error signals for the two loops, until the new . is dened steady state is reached at as the maximum settling time of all loops (See Fig. 2). The incremental equation from the second steady state to the new state becomes (A1a)

Fig. 10.

CCS under automatic control.

in automatic mode, the main steam pressure loop and the power output loop have had to be controlled by operators because of the limitations of the conventional CCS. As it is difcult to guarantee safe operation of the plant using a conventional CCS, such studies were not conducted, and, hence, no comparative results are available between the proposed approach and the conventional CSS. Hence, the results of the proposed approach and manual operation of the plant is presented, to demonstrate the improvement in performance with the proposed CSS. VII. CONCLUSION A new coordinated control strategy for the boiler-turbine unit in a power plant was proposed in this paper to improve the system performance under the load following mode. The system consists of two levels with a fuzzy inference system for autotuning the PID controller. The CCS and fuzzy autotuning PID controllers have been implemented in a 300 MW boiler-turbine unit in China for more than two years. The system has been performing very well after three years of ne-tuning. In summary, we have adverted the following: 1) the CCS had successfully replaced the manual operation in loop coordinate control and the performance has been very robust;

952

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 13, NO. 6, NOVEMBER 2005

(A1b) from to , while keeping 2) Make a step change in as before, record the error signals for the two loops, until the new steady state is arrived at . Again, the incremental equation from the third steady state to the new state can be written as (A2a) (A2b) Combine (A1a), (A1b), (A2a) and (A2b) into matrix form (A3a) where

decentralized controllers are proportional controllers, that is and are constants, the relation between the original both system inputoutput and the decentralized identication system are given as. Consider a rst-order plus time delay and a second-order plus 1 and 2 are given time delay system, the solutions for as follows. 1, each loop output results becomes 1)

(A5) which can be written into the compact form

where

2)

2, (A5) becomes

and

(A3b) Because of thus, the matrix is nonsingular if . From (A1) and (A2), it can be proved that , It is again expressed as (A6)

where

Therefore, for a bound-inputbound-output (BIBO) , in Laplace domain. stable system, Now that is nonsingular, can be solved by (A4) Substituting (A1b), (A2b) and (A3b) into (A4), we obtain the equation shown at the bottom of the page. Consequently, the problem of identication of coupled closed-loop MIMO system is transformed into the identication of four single open-loop problems. Suppose that Equations (A5) and (A6) can be solved by the Least Squares methods for each transfer functions 1, 2 1, 2, to form the regression form and

LI et al.: A NEW COORDINATED CONTROL STRATEGY FOR BOILER-TURBINE SYSTEM

953

ACKNOWLEDGMENT The authors would like to thank the anonymous reviewers for their helpful comments and constructive suggestions with regard to this paper. REFERENCES
[1] C. Maffezzoni, Boiler-turbine dynamics in power-plant control, Control Eng. Pract., vol. 5, no. 3, pp. 301312, 1997. [2] S. Goto, M. Nakamura, and S. Matsumura, Automatic realization of human experience for controlling variable-pressure boilers, Control Eng. Pract., vol. 10, no. 1, pp. 1522, 2002. [3] T. Chai, H. Yue, and Y. Bai, Intelligent control of coal-pulverizing systems with ball-tube mill, in Proc. IFAC 14th Triennial World Congr., Beijing, China, 1999, pp. 97102. [4] P. Zgorzelski, H. Unbehauen, and A. Niederlinski, A new simple decentralized adaptive multivariable regulators and its application to multivariable plants, in Proc. IFAC 11th World Congr., vol. 4, Tallinn, 1990, pp. 226231. [5] H. Liu, A Study on advanced control techniques application in the large thermal power generating unit, Doctoral dissertation, AU: WHAT DEPARTMENT? Northeastern Univ., China, 2000. [6] J. A. Rossiter, B. Kouvaritakis, and R. M. Dunnett, Application of generalized predictive control to a boiler-turbine unite for electricity generation, Inst. Elect. Eng. Proc.D, vol. 138, no. 1, pp. 5967, 1991. [7] B. W. Hogg and N. M. El-Rabie, Generalized predictive control of steam pressure in a drum boiler, IEEE Trans. Energy Convers., vol. 5, no. 3, pp. 485492, Sep. 1990. , Multivariable generalized predictive control of a boiler system, [8] IEEE Trans. Energy Convers., vol. 6, no. 2, pp. 282288, Jun. 1991. [9] W. Wang, H. X. Li, and J. T. Zhang, Intelligence-based hybrid control for power plant boiler, IEEE Trans. Contr. Syst. Technol., vol. 10, no. 2, pp. 280287, Mar. 2002. [10] G. Zhang, C. Shao, and T. Chai, A new method for independently tuning PID parameters, in Proc. 35th IEEE Conf. Decision Control, Kobe, Japan, 1996, pp. 25272532. [11] Q. Bi, W. J. Cai, and Q. G. Wang et al., Robust identication of rstorder plus dead-time model from step response, Control Eng. Pract., vol. 7, no. 1, pp. 7177, 1999. [12] , Advanced controller auto-tuning and its application in HVAC systems, Control Eng. Pract., vol. 8, no. 5, pp. 633644, 2000. [13] E. Poulin, A. Pomerleau, and A. Desbiens et al., Development and evaluation of an auto-tuning and adaptive PID controller, Automatica, vol. 32, no. 1, pp. 7182, 1996. [14] Q. G. Wang, C. C. Hang, and Q. Bi, Process frequency response estimation from relay feedback, Control Eng. Pract., vol. 5, no. 9, pp. 12931302, 1997. [15] R. Bandyopadhyay and D. Patranabis, A fuzzy logic based PI autotuner, ISA Trans., vol. 37, pp. 227235, 1998. [16] K. K. Tang, Q. G. Wang, and C. C. Hang, Advances in PID Control. New Yprk: Springer-Verlag, 1999. [17] X. Liu, T. Chai, and H. Liu, Fuzzy logic strategy for boiler control, Acta Automatica Sinica, vol. 24, no. 4, pp. 534538, 1998. [18] T. Sudkamp and R. J. Hammel, Interpolation, completion, and learning fuzzy rules, IEEE Trans. Syst., Man, Cybern., vol. 24, no. 2, pp. 332342, Feb. 1994. [19] W. K. Ho, C. C. Hang, W. Wojsznis, and Q. H. Tao, Frequency domain approach to self-tuning PID control, Control Eng. Pract., vol. 4, no. 6, pp. 807813, 1996.

where . Its Least Squares estimation for are

(A7) are found from (A7), Once ered from , and can be recov-

for

, and

and

from

for

2, respectively.

APPENDIX ZIEGLERNICHOLS FREQUENCY RESPONSE METHOD [16] The ZieglerNichols frequency method is based on using the controller connected as a proportional controller, the experiment is carried out in the following procedures. 1) Connect a controller to the process, set the parameters , and so that control action is proportional, i.e., 0. 2) Increase the control gain slowly until the process starts to oscillate. , and the period of the 3) The gain when this occurs is oscillation is . 4) Determine the PID controller parameters according to the ZieglerNichols method as shown in the gure at the top of the page.

Shaoyuan Li (SM05) was born in 1965. He received the B.S. and M.S. degrees from Hebei University of Technology, Tangshan, China, in 1987 and 1992, respectively, and the Ph.D. degree from the Department of Computer and System Science, Nankai University, Tianjin, Beijing, China in 1997. He is currently a Professor at the Institute of Automation, Shanghai Jiao Tong University, Shanghai, China. His research interests include fuzzy systems and nonlinear system control.

954

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 13, NO. 6, NOVEMBER 2005

Hongbo Liu was born in 1964. He received the Ph.D. degree from the Research Center of Automation, Northeastern University, China, in 2000. He is currently working toward the Ph.D. degree at the Institute of Automation, Shanghai Jiao Tong University, Shanghai, China. His research interests include complex thermal process modeling and control, fuzzy control, and adaptive control.

Yeng-Chai Soh received the B.Eng. degree in electrical and electronic engineering from the University of Canterbury, New Zealand, in 1983 and the Ph.D. degree in electronic engineering from the University of Newcastle, Australia, in 1987. He is currently a Professor in the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore. Since 1995, he has been the Head of the Control and Instrumentation Division. His current research interests are in the areas of robust system theory and applications.

Wen-Jian Cai was born in 1957. He received the B.S. and M.S. degrees from Harbin Institute of Technology, Harbin, China, in 1980 and 1983, respectively, and the Ph.D. degree in systems engineering, Oakland University, Rochester, MI, in 1992. He is currently an Associate Professor in the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore. His research interest includes advanced process control, fuzzy logic control, robust control, and estimation techniques.

Li-Hua Xie received the B.E. and M.E. degree in electrical engineering from the Nanjing University of Science and Technology, Nanjing, China, in 1983 and 1986, respectively, and the Ph.D. degree in electronic engineering from the University of Newcastle, Australia, in 1992. He is currently an Associate Professor in the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore. His current research interests include optimal and robust control.