steam power plant configuration, design, and...

Advanced Review

Steam power plant configuration,design, and controlXiao Wu,1 Jiong Shen,1 Yiguo Li1 and Kwang Y. Lee2∗

This article provides an overview of fossil-fuel power plant (FFPP) configura-tion, design and especially, the control technology, both the conventional and theadvanced technologies. First, a brief introduction of FFPP fundamentals and con-figurations are presented, followed by the description of conventional PID-basedcontrol system in the FFPPs and its short-comings. As the major part of this writing,different advanced control strategies and applications are reported, with their sig-nificant features outlined and discussed. These new technologies are collected fromboth the academic studies and industrial practices, which can improve the perfor-mance of the FFPP control system for more economic and safe plant operation.The final section presents a view of the next generation FFPP control technologies,emphasizing potential business and research opportunities. © 2015 John Wiley & Sons,Ltd.

How to cite this article:WIREs Energy Environ 2015. doi: 10.1002/wene.161

INTRODUCTION

Fossil fuelled power plant (FFPP) refers to a groupof power generation devices that convert the chem-

ical energy stored in the fossil fuel such as coal, gas,oil into thermal energy, mechanical energy and finallyelectrical energy. In the past hundred years, FFPPs arethe most widely used facilities in the power industryand play a fundamental role in social production andlife. According to the 2013 Key World Energy Statis-tics published by the International Energy Agency(IEA), in 2011, the annual generation of electric-ity from all types of sources was 22,126 TWh andFFPPs provided 15,054 TWh, accounted for 68% ofthe total electricity generation. Although the rapidincrease of global energy crisis, combined with theconcerns about environment issues has led to an exten-sive promotion of nuclear and renewable energy, forthe most parts of the world, the trend of conven-tional fossil-fuel-dominated electric power generation

∗Correspondence to: [email protected] Laboratory of Energy Thermal Conversion and Controlof Ministry of Education, School of Energy and Environment,Southeast University, Nanjing, China2Department of Electrical and Computer Engineering, Baylor Uni-versity, Waco, TX, USA

Conflict of interest: The authors have declared no conflicts of interestfor this article.

will not change in a foreseeable future. For this reason,developing and operating the FFPPs according to themost suitable available technologies are very impor-tant and should be the most effective and direct wayto save energy and reduce the pollution.

The history of FFPP can be traced back to the late19th century, the simple D.C. generators were coupledto coal-fired, reciprocating piston steam engines, pro-ducing electricity primarily for district lighting. Theseinitial plants typically operated at low temperatureand pressure conditions (150∘C, 0.9 Mpa) and couldonly generate 30 kw electricity. Through a century’stechnological developments, power plants have nowbeen evolved into a highly complex system that canoperate at supercritical conditions of 28.5 Mpa and600∘C, generating 1300 MW of electricity with muchhigher efficiency. Although there are many variationsin power plant configuration and design, the basicworking principle of the FFPPs keeps the same: fossilfuel is combusted, generating high pressure and tem-perature steam, which is then expanded to rotate a tur-bine, and drives the generator to produce electricity.1

For the FFPP, the main task of the controlsystem is to regulate the electrical power outputto meet the demand of the grid while maintain-ing the main thermal dynamical variables such assuperheater/reheater steam temperature, throttle pres-sure, furnace pressure, drum water level, within

© 2015 John Wiley & Sons, Ltd.

Advanced Review wires.wiley.com/wene

given tolerances to keep the power plant operatingsafely. Generally, such an objective is achieved viathe multi-loop proportional-integral-derivative (PID)based controllers. The approach has been proven tobe highly reliable and can attain a satisfactory per-formance under normal operation maintained at baseload, where plant characteristics become almost con-stant.

However, during the past few decades, powerindustry has undergone some significant changes,and as the primary devices of power production,FFPPs have been endowed with higher operationalrequirements:

1. The growth of electric power demand increasesthe magnitude of the cyclic variation of the gridload, and the renewable sources, such as wind,solar and hydro power, are severely influencedby the season and the weather condition; thus,the FFPPs have to participate in the grid powerregulation frequently and respond to the loaddemand variation quickly within a wide oper-ation range.

2. The privatization and deregulation of theelectricity industry has changed the power gen-eration business from a cost-plus, monopolyenvironment with an obligation to serve, toa competitive environment for the sale of itsproduct. For this reason, power plants areincreasing in size and becoming more complexin order to achieve high efficiency and the scaleof economy. Furthermore, the aforementionedthermal parameters should be more stringentlycontrolled so that the plant can operate in anoptimal mode at all times.

Therefore, control problems to deal with issues,such as nonlinearity over a wide operation range, largeinertial and time varying behavior, and strong cou-pling among the multitude of variables, become severein the FFPPs. Consequently, the conventional PI/PIDbased controllers are no longer sufficient in meetingperformance specifications, even if they are well tunedat a given load level. On the other hand, with the helpof modern computer and instrumentation techniques,utilizing Distributed Control System (DCS) is now theroutine rather than the exception, which makes theimplementation of advanced controllers possible inthe FFPPs. The primary purpose of this writing is topresent an updated, representative snapshot of vari-ous control strategies that are being applied to theFFPPs and describe how they can help in improvingthe quality and performance of plant operation. The

information reported here are collected from both aca-demic researches and engineering practices.

A brief introduction of FFPP fundamentals andconfigurations are presented first, followed by thedescription of conventional PID-based control systemin the FFPPs and the associated problems. As amajor part of this writing, different advanced controlstrategies and applications are reported, with theirsignificant features outlined and discussed. The finalsection presents a view of the next generation FFPPcontrol technologies, emphasizing potential businessand research opportunities.

PLANT CONFIGURATION AND DESIGN

The essence of power production process in all types ofthe FFPPs is energy conversion. In the vast majority ofthe FFPPs worldwide, water/steam is commonly usedas the working fluid, which is alternately vaporizedand condensed in a closed circuit following a thermaldynamic cycle. Within this cycle, the chemical energyof the fossil fuel is transformed into steam thermalenergy by the boiler, then it is transformed intorotational mechanical energy by the turbine, andfinally it is transformed into electric energy by thegenerator. This kind of FFPPs is also called steampower plant, and depending on the operating steampressure, it can be classified into subcritical plants andsupercritical plants.

Subcritical Steam PlantIn subcritical power plants, the steam parame-ters never exceed the critical point: 22.115 Mpa,374.12∘C. Because under this critical point, liquidwater must go through a vaporization stage to becomesteam; in most of the large-scale subcritical plants,drums are typically utilized to separate the steam outof the boilers.

Figure 1 provides a simplified illustration of acoal-fired subcritical power plant, which is comprisedof two basic systems: the fuel/air-flue gas system andthe water-steam system.2–4

The fuel/air-flue gas system is also called the fire-side of the plant. In this system, the raw coal is trans-ported to the coal hopper by the conveyor and entersthe pulverizing mill; where grinding and crushing takeplace. The qualified (smaller and lighter) coal parti-cles are then separated and entrained in the air flow,and carried into the burner. Finally, the combustionoccurs in the furnace, generating high temperature(above 1000∘C) flue gases. The air needed for com-bustion is delivered to the furnace and mill by theforced draught fans and an air preheater is installed in


WIREs Energy and Environment Steam power plant configuration, design, and control

Key1. Cooling tower 8. Condenser 15. Coal hopper 22. Air intake

2726

25

23

22

211917

14

2420181613

15

12

11

10965

1

2

43

8

7

2. Cooling water pump 9. Intermediate pressure turbine 16. Pulverized fuel mill 23. Economizer

3. Pylon (termination tower) 10. Steam governor 17. Boiler drum 24. Air preheater

4. Unit transformer 11. High pressure turbine 18. Ash hopper 25. Precipitator

5. Generator 12. Deaerator 19. Superheater 26. Induced draught fan

6. Low pressure turbine 13. Feed heater 20. Forced draught fan 27. Chimney stack

7. Boiler feed pump 14. Coal conveyor 21. Reheater

FIGURE 1 | Simplified coal-fired subcritical power plant (Picture from http://en.wikipedia.org/wiki/Thermal_power_station#Boiler_furnace_and_steam_drum).

the path to warm the air being fed utilizing the heatremaining in the exhaust gases leaving the furnace,through which the efficiency of the combustion can beimproved. The objective of this process is convertingthe chemical energy in the fuel into the thermal energyin the flue gases. The flue gases in the furnace trans-fer the heat to the water wall by radiation and thenflow through multiple stages of superheaters, whichare suspended on the horizontal passage at the topof the furnace. Depending on the installed positions,some superheaters are radiant type, which absorb heatby radiation; others are convection type, absorbingheat from fluid; some are a combination of the twotypes. Through either type, the extreme heat in theflue gases is transfered to the superheater piping andthe steam within. After leaving the superheater, the fluegases pass over reheater, economizer and air preheater,where almost all of their remaining heat is extractedto reheat the steam or prewarm the feed-water andfeed-air. The induced draught fans work in conjunc-tion with the forced draught fans, then pull the fluegases into the precipitator, and finally out of the boiler

through the chimney. The falling slags and ashes arecollected in the ash hopper and delivered to the ashsystem of the plant.2–4

Water-steam system is also referred to as thewaterside of the plant, which operates following theRankine cycle. The procedure within this systembegins with the feedwater being drawn from the con-denser and delivered to the boiler by the feed pumps.To improve the plant efficiency, a series of low andhigh pressure feed heaters and an economizer are uti-lized to heat the feedwater with the steam bled fromthe turbine and the remaining heat of the flue gases.The deaerator is also installed in this path to removethe dissolved gases in the feedwater by vigorously boil-ing and agitating it. The drum supplies the feedwaterto the waterwall of the furnace to absorb the radiationheat and separate the resulting saturated steam fromthe incoming saturated feedwater. The steam is thenfurther heated through multiple stages of superheatersto reach higher temperature and pressure. Finally, thesteam expands along the turbines and rotates themto a given high speed (3000/3600 rpm), which then


http://en.wikipedia.org/wiki/Thermal&uscore;power&uscore;station#Boiler&uscore;furnace&uscore;and&uscore;steam&uscore;drum


Carbon

capture

Flue gas

desulpherization

(FGD) unit

Feed

water

pump

Feed water

heater

Feed water

heater

Feed water

booster pumpDeaerator

Condensate

pumpCondenser

cooling water

pump

Cooling tower

Cooling water

retureCondenser

Generator

Steam turbineSteam turbine

Main steam

valve

Attemperator

Valve

Boiler

Superheater

Reheater

Reheat stop

valve

FIGURE 2 | Simplified supercritical power plant (Picture fromhttp://www.flowserve.com/Industries/Power-Generation/Conventional-Steam/Flowserve-Products-Used-in-Supercritical-Units,en_US).

drive a generator to produce electricity. Usually, thereare multiple stages of turbines, and for a higher plantefficiency, after the expansion in the high pressureturbine, the steam is extracted and reheated in theboiler, and then delivered to the medium and low pres-sure turbines. The saturated steam leaving the lowpressure turbine is condensed back into liquid in thecondenser.2–4

Supercritical Steam PlantIn contrast to the subcritical plant, supercritical plantis another type of steam plant, where the steamgenerator operates at pressure greater than the criticalpoint, 22.115 Mpa. Because above such a pressure, thephysical turbulence that characterizes boiling ceasesto occur, and instead, the liquid water immediatelybecomes steam once is heated above the criticaltemperature (374.12∘C). Therefore, the drum used inthe subcritical plant, where the evaporation separationprocess occurs can be completely eliminated, and thefeedwater circulates only once in the furnace in eachcycle (Figure 2).2 For this reason, the ‘once throughsteam generator’ is designed and employed in allsupercritical plant.4

Current Status of the Steam Power PlantThe subcritical plant is still expected to remainthe main choice in some countries due to its sim-plicity in operation and control, belief in higher

reliability and low technical risk. However, thesupercritical/ultra-supercritical plant is now greatlypromoted, because operating the plant at highertemperature and pressure can increase its efficiency,potentially lowering the amount of fossil fuel con-sumed and the emissions generated.

Currently, there are more than 600 supercriticaland ultra-supercritical power plants with total capac-ity above 400 GW in the world (status 2010, Figure 3).These supercritical plants can achieve efficiencies ofmore than 42%, compared with subcritical plants’33%–39%. According to the USA DOE power gener-ation initiative: Vision 21, by the year 2020, the steamin the ultra-supercritical power plants is expected toreach 760∘C and 38.5 Mpa, which will enhance theplant efficiency to more than 50%.

In spite of the great advantages of the super-critical plants, there are still barriers for building thistype of the plant: the high thermal stresses and fatiguecracking in the critical sections of the plants as well asthe resulting lower reliability and higher maintenancecosts. Thus an identifying, evaluating, and qualifyingpotential alloy material is the major challenge for thesuccessful implementation of supercritical technology.

CLASSICAL CONTROL OF THE FFPPAs stated previously, the FFPPs, especially the steampower plants, are complex, multivariable, and interac-tive processes, thus a well-designed control system is


http://www.flowserve.com/Industries/Power-Generation/Conventional-Steam/Flowserve-Products-Used-in-Supercritical-Units,en&uscore;US


800

Capacity (

GW

)

700

600

500

400

300

200

100

Ultra-supercritical Supercritical Subcritical

Austr

alia

Germ

any

Japan

Kore

a

Russia

South

Afr

ica

United

Kin

gdom

China India United States

2010

2012

2014

2010

2012

2014

2010

2012

2014

0

FIGURE 3 | Capacity of supercritical and ultra-supercritical plant in major countries (refers to capacity in 2010 unless specified otherwise, Picturefrom http://www.iea.org).

required in the plants to ensure the correct operationof the entire process, i.e., rapidly following the gridload demand and controlling relevant process vari-ables such as: throttle pressure, superheater/reheatersteam temperature, furnace pressure, drum water level(subcritical plant), etc., so that high efficiency, durabil-ity and safety can be attained in the plant.

The boiler-turbine unit control schemes havegone through several decades of evolution and, typ-ically, a cascade of PI/PID controllers based onsingle-input single-output control loops is designedin the plant to fulfill such tasks.6,7 The remain-der of this section will focus mainly on the con-ventional boiler-turbine coordinated control system(CCS), steam temperature control system, combus-tion control system, and feedwater control system, inwhich the respective thermal dynamic variables arecontrolled separately.

Boiler-Turbine Coordinated Control SystemCurrent plant or unit control strategies allow gener-ation of the grid load demand while maintaining thebalance among the process variables within the unit.Mainly, they match the boiler steam flow energy out-put to the energy required by the turbine-generator tomatch the unit load demand at all times.6 The coor-dinated control system (CCS) constitutes the upper-most layer of the control system, and it is responsiblefor driving the boiler-turbine-generator set as a sin-gle entity, harmonizing the slow response of the boilerwith the faster response of the turbine, to achievefast and stable unit response during load trackingmaneuvers and load disturbances.

For the FFPP, power output and throttle pres-sure are the two most important variables. Exter-nally, the power output reflects a balance between theplant’s power generation and grid’s power demand;internally, the throttle pressure naturally represents abalance between the boiler’s energy supply and tur-bine’s energy need. The dominant behavior of the unitis governed through the power and pressure controlloops. Therefore, the central task of the CCS is to reg-ulate the power output to meet the demand of thegrid while maintaining the throttle pressure within agiven tolerance. Evolved from multiple single-inputsingle-output control loop (decentralized) configura-tions based on PID control algorithms, currently, thereare two possible modes for coordinated control: coor-dinated boiler-following (BF) mode and coordinatedturbine-following (TF) mode.1,5,7,8

Historically, boiler following schemes werethe first to be used.9 In boiler following mode, theboiler awaits the actions of the turbine to matchthe requested generation. The turbine control valvesregulate the steam flow into the turbine in terms ofthe power demand. Then, the boiler controls respondto the changes in steam flow and pressure. The basicprinciple of the coordinated BF mode is illustratedin Figure 4. The advantage of this approach is a fastresponse to load changes, nevertheless, it should benoted that such rapid response is basically achievedby using the stored thermal energy in the plant, thusit is effective only for a small demand change. Thedisadvantage of the coordinated BF mode is that, in itspure form, this approach shows a less stable throttlepressure control since the boiler has a tendency to


http://www.iea.org


Boiler main

control

Turbine main

control

Turbine sub-

control

system

Governor valve

μTTurbine

~

Boiler

Fire rate μB

Boiler sub-

control

system+

–

P0

PT

+

–

E0

E

BD

TD

Generator

FIGURE 4 | Working Principle of the Coordinated BF mode (E0: load set-point; P0: main steam pressure set-point; E: power output; PT: mainsteam pressure output; BD: boiler demand; TD: turbine demand).

overshoot because it requires some time to match theturbine.10

Turbine following began around the late 1960sand early 1970s.9 In the coordinated TF mode ofcontrol, the turbine follows the actions of the boiler tomatch the requested generation. The power demandis used by the combustion control at the boiler toadjust the fuel and air into the furnace to modify thesteam production. Then the turbine controls respondby adjusting the throttle valve openings to keep thepressure at the setpoint value. The advantage of thisapproach is its very stable response to load changeswith minimal steam pressure fluctuations. The maindisadvantage is that this approach does not make useof the energy storage capability of the boiler, thusproducing a rather slow response.10 For this reason, itis mainly used for a large base-load plant or a gas-firedplant, which has a relatively quick response comparedto the coal plant.

It is worth mentioning that, to improve theperformance of the CCS in both BF and TF modes,the power demand is fed both to the boiler system (BF)and turbine system (TF)9 (this is shown as dotted-linein Figures 4 and 5), so that the large inertial behaviorof boiler can be partly compensated (in BF) and theturbine’s ability to respond quickly can be utilized (inTF). This coordinated control scheme is now widelyused in practice.

However, the PI/ PID control systems, whichare based on a cascade of separate SISO loops,cannot fully account for the interactions among the

different process variables in the nonlinear multi-inputmulti-output (MIMO) power plant. Therefore, it isstill very difficult for the classical CCS to achievea satisfactory control performance in both poweroutput and throttle pressure.

For this reason, various advanced modeling andmultivariable control technologies are studied, aimingto realize a real coordinated control of boiler-turbine.This will be introduced in the next section.

Combustion Control SystemUnder the CCS, the mission of the combustion systemis to provide enough thermal energy while guarantee-ing the efficient and safe operation of the boiler. Suchrequirements are fulfilled by controlling relevant vari-ables in the plant, namely:

(a) Fuel flow rate to maintain throttle pressure (inBF mode) or electrical power output (in TFmode);

(b) Excess air coefficient or optimal oxygen contentin the flue gases to ensure appropriate air flowrate;

(c) Furnace pressure to guarantee the safety of theboiler.

The regulation of the above three variable canbe achieved through the manipulation of fuel feed-ers, forced draught (FD) dampers and induced draught



Boiler main

control

Turbine main

control

Turbine sub-

control

system

Governor valve

μT

Boiler sub-

control

system

E0

Turbine

~

Boiler

Fire rate μB

+

–

P0

PT

+

– E

BD

TD

Generator

FIGURE 5 | Working Principle of the Coordinated TF mode (E0: load set-point; P0: main steam pressure set-point; E: power output; PT: mainsteam pressure output; BD: boiler demand; TD: turbine demand).

Boiler main control

Fuel

regulator

Air

regulator

Pressure

regulator

BD

Coordination level

FIGURE 6 | Combustion control system (BD: boiler demand).

(ID) dampers, respectively, resulting in three indivisi-ble sub-regulators, i.e., the fuel regulator, air regulatorand pressure regulator as shown in Figure 6, whichform the combustion control system.3,5

The task of the fuel regulator is to provideenough fuel to the boiler so that the generated thermalenergy can exactly match the load demand. A cascadecontrol strategy is usually employed in this system,which is shown in the left part of Figure 7 for a BFmode plant. Because the fuel flow rate, especially forthe coal flow, is difficult to measure, and the feederspeed signal cannot reflect the variation of the heatingvalue of the fuel, in most of the FFPPs, the heat

quantity signal M is used in the inner loop to rapidlydeal with any disturbances due to the variation of thefuel in heating values. The heat quantity signal can beestimated through the equation

M = D + Cbdpb∕dt, (1)

where, D is the steam flow rate representing the heatabsorbed by the working substance, and Pb is the drumpressure representing the heat stored in the boiler, Cbis the heat storing coefficient of the boiler.

The air regulator is used to guarantee the effi-ciency of the combustion; to be specific, guarantee asuitable ratio between the amounts of fuel and airbeing supplied to the furnace. An undersupply of theair will prevent the fuel from complete burning; in con-trast an oversupply of the air will absorb heat and thusincrease the heat waste in the exhaust gas. In practice,a certain amount of excess air is needed rather thankeeping the fuel/air ratio at the stoichiometric value.

Since the fuel flow rate has already been deter-mined by the fuel regulator, it is direct to design a ratiocontroller to keep the air flow rate matching the fuelflow rate. However, considering that the heating valueof the fuel can vary from time to time, it is a challengeto set the ratio co-efficient in operation; therefore, theoxygen content in the flue gas is measured which canreflect the combustion condition directly, regardless ofthe fuel variation.



FIGURE 7 | Three sub regulators of boilercombustion control system (P0: main steam pressureset-point; PT: main steam pressure output; D: mainsteam flow rate; O2: oxygen content in the flue gas; V:air flow rate; Pss: furnace pressure set-point; Ps:furnace pressure output; BD: boiler demand).

+

f1(x)

PI5

f2(x)

PI1

PI2 PI3

f3(x)

PI4

pTp0

BD

– +

+ –

M

+ –

V

+ – +

ps pss

DO2

×

–

Fuel regulator Air regulator Pressure regulator

A classic air regulator using the oxygen contentsignal is illustrated in the middle part of Figure 7. It isa cascade control system. The inner loop receives thefuel signal and air signal to attain a basic fuel/air ratioand the outer loop receives the oxygen content signaland compare it with the optimal oxygen content value(which is calculated offline for a given load) to achievea tight control of the air input.

The task of the pressure regulator is to main-tain the furnace pressure close to the atmosphericpressure; so that the hazardous gases escaping andcool air entering the boiler can be prevented. Usually,furnace pressure is required to be maintained at20–50 Pa below the atmospheric pressure. Such a taskis achieved by the use of feedforward-feedbackcontrol system as shown in the right part ofFigure 7.

Feedwater Control SystemThe objective of the feedwater control is to supplyenough water to the boiler to match the evaporationrate. For the subcritical plant, because the separationof steam from water always happens at the drum,maintaining the drum level naturally represents thebalance between the feedwater supply and steamgeneration.

Controlling the drum water level within a giventolerance is important for the safe operation of theplant: a high level will increase the risk of water beingcarried over into the steam circle, which may notonly lead to a fluctuation of steam temperature, butalso cause fouling and damage of the superheaters;conversely, a low level will cause the waterwall pipingto be damaged from insufficient cooling. Both canresult in catastrophic failures.

The drum water level is determined by both thevolume of the water in the drum and the volume ofthe steam bubbles under the water level. Thus, thedrum water level can be influenced by feedwater flowrate, steam flow rate, heat quantity generated fromcombustion and many other variables, and its controlpresents a complex problem due to the large inertiabehavior of these disturbances and a ‘swell and shrink’effect.3

For these reasons, a three-element cascade con-troller is typically used in the plant, which is illustratedin Figure 8. The steam flow rate signal D is used asthe feedforward signal; such a design can make thefeedwater flow rate respond quickly to the variation ofthe steam flow rate, thus avoid the effect of ‘swell andshrink’. The feedwater flow rate signal is used to formthe inner-loop of the control system and a secondarycontroller is designed for a quick rejection of the dis-turbance inside the feedwater system. The drum waterlevel H is finally fed back to the primary controller foran accurate regulation.3,5

The main difference between a supercriticalplant and subcritical plant is at the water-steam sys-tem. For the supercritical plant, feedwater circulatesonly once in the furnace in each cycle and there is noclear disengagement surface between steam and water.However, both the fuel flow rate and feedwater flowrate can greatly influence the position of the surface, ifsuch a surface deviates far away from a designed level,the steam temperature in the superheater would alsohave a deviation far away from the set-point. There-fore, generally, a ratio controller is designed for thesupercritical plant to regulate the feedwater flow rate,keeping it matching the fuel flow rate, so that thesteam temperature/enthalpy out of the separator canbe controlled within the given range.



Δp

PID1

Δp

aD

PID2

aW

KZ

Δp

Drum

Economizer

Super heater Steam flow D

Drum water

level H

FIGURE 8 | Three elements cascade feedwater control system for adrum type boiler (Δp is the differential pressure transmitter;

√is the

square root extractor; aw, aD are the sensitive coefficients of feedwaterflow rate and steam flow rate signals; Kz is the actuator of thefeedwater flow control valve).

Steam Temperature Control SystemSuperheater steam temperature (SST) and Reheatersteam temperature (RST) are two of the most criticalvariables to be controlled in a steam power plant.They must be tightly controlled within a small range,as shown in Figure 9, for the following safety andeconomy reasons:

1. Excessively high temperature will lead to mate-rial damage on the superheater/reheater steampipes at the inlet of the turbine;

2. Lower temperature will reduce the efficiency ofthe plant, moreover, it will build up the steamhumidity in the rear of the low pressure turbinethat would erode the turbine blades; and

3. Large temperature variation will increase thethermal stress of the piping material and mag-nify the variations of the air gap between rotorand stator of the turbine, thus threatening thesafety of the plant.

For the subcritical plant, there are many factorsthat will influence the SST: mostly the rate of the

T_sp

T_sp+5

T_sp+10

T_sp-10

T_sp-5

Temporary operation region

Long term operation region

Long term operation region

Temporary operation region

Dangerous operation region

Low efficiency operation region

FIGURE 9 | Operation regions of superheater steam temperature(T_sp is the temperature set-point and the numbers representtemperature deviations in Celsius degree).

steam flow, heat transfer from flue gas and the rateof spray water flow. The first two have a relativelyquick influence on the SST while the spray water’sinfluence is quite slow. However, during the operation,the steam flow rate has to match the load demand;the regulation of heat transfers (for example, theburner tilt or the gas recirculation) will influence theefficiency and security of combustion. Therefore, thespray water flow becomes the only variable to controlthe SST and the cascade control system is generallyemployed, which uses an inner loop to handle the largeinertia property.

A classical SST control system is illustrated inFigure 10. The inner loop receives the steam temper-ature signal T2 immediately after the attemperation,which is required to reject the temperature distur-bances originating upstream as well as the self distur-bance in the spray water. The inner loop is, of course,much faster than the outer loop, which receives thefinal steam temperature signal T1 to achieve an accu-rate control performance.2,5

For the supercritical plant, as analyzed in section3.3, the regulation capability for the spray water isvery limited, and the superheater steam temperatureis mainly controlled by regulating the fuel/feedwaterratio. The steam temperature/enthalpy out of theseparator is first controlled by keeping the feedwaterflow rate matching the fuel flow rate; spray water isthen used for a tighter control of the superheater steamtemperature.

The control of the reheater steam temperature ismainly attained by regulating the dampers that controlthe flow of the flue gases across the reheater tubebanks. The spray water is only used in emergencycase, because it will reduce the amount of steam whichexpands in the high pressure turbine and will reducethe efficiency of the whole plant.



γT2

PI2

KZ

γT1

PI1

Steam

Superheater I Superheater II

Attemperator

valve

T2 T1

Attemperator

FIGURE 10 | Classical cascade SST control in the FFPP (T1 is finalsteam temperature; T2 is the steam temperature signal immediatelyafter the attemperation; 𝛾T1 and 𝛾T2 are the temperature transmitters;Kz is the actuator of the attemperator valve).

The Drawbacks of the Conventional ControlThe conventional PI/PID based control system hasbeen successful in FFPPs and are very reliable. Such adesign has also been proved for its value for regulationand disturbance rejection during normal operationmaintained at/around a base load. However, as theFFPPs increase in size and participate in grid loadregulation more frequently, control of FFPP unit hasbeen shown to be a challenge, due to severe non-linearity in the multitude of variables over a wideoperation range, tight operating constraints and largeinertia behavior. Moreover, the equipment wear, envi-ronmental changes, fuel variation, etc., will result insignificant disturbances and plant behavior variations.Consequently, the conventional PI/PID strategiesare no longer sufficient in meeting performancespecifications because of the following drawbacks:

(a) The main drawback of the PI/PID controlsystems, based on separate single-input,single-output (SISO) loops, is that they donot account for the interactions of the differentthermal properties in the plant.

(b) In general, the PI/PID controller parameters areoptimized at a given operating point and thenfixed. Therefore, when wide-range load follow-ing is required for the FFPPs, the performanceof the plant operation is decreased because thenonlinearity becomes significant.

(c) The PI/PID controllers are not possible to han-dle the constraints of manipulated variablesin the controller calculation stage, thus even

when the controller parameters are well tuned,the performance is still decreased when phys-ical limitations (both magnitude and rate) ofthe valves are involved. This may also causethe integral windup when a sharp change of thepower demand occurs.

Therefore, various advanced control strategieshave been proposed in both academic studies andindustrial practices, aiming at improving the per-formance of the FFPP control system for economicand safe plant operation. A migration from classicalPI/PID based control system to new concepts based onadvanced control techniques in FFPPs will take placein a foreseeable future. The next section introducesfour types of different control strategies with diverseapplications in the FFPPs.

ADVANCED CONTROL OF THE FFPP

Advanced PI/PID ControlThe advanced PI/PID control refers a class of methodswhich implement the state-of-the-art design or tun-ing technologies on the basis of conventional PI/PIDcontrol loops. Because the PI/PID based control sys-tem has already been widely accepted and employedthroughout the FFPPs, such a method can directly andeffectively improve the operation performance with-out altering the original simple structure, operationprocedures and concepts which are well understood byplant engineers and operators. Recently, auto-tuningand gain scheduling PI/PID controls are two exten-sively studied methods in power plants.

Auto-Tuning PI/PID ControlAs stated before, the power plant control system con-sists of many SISO PI/PID loops which are stronglyinteracting to each other. The most common methodof tuning these controllers in the FFPPs is the socalled ‘trial and error’ method.11 It generally tunesthe control loops sequentially, beginning from theone with least interaction. For complex MIMO sys-tem like FFPP, such a method requires considerableexpertise and experience and may still be difficult toattain a satisfactory overall performance. An alter-native method is to design an analytical compensa-tion of control-loop interactions for wide-range plantoperation.12

Auto-tuning PI/PID control can potentially han-dle the interactions among process variables andloops, where an objective function reflecting an overalldynamic control performance is utilized and throughminimization of this objective function, all controller



FFPP

Model

Heuristic

search

Objective

function

Candidate controller

parameters

Search

mode

Run

mode

PID controller

Update

Reference Plant

output

FIGURE 11 | Block diagram of a typicaltuning method for PI/PID parameters.

parameters can be determined simultaneously andoptimally.

Figure 11 shows the block diagram of a typicaltuning method in Refs 13–15, in which a dynamicmodel of the plant is used to test different groupsof controller parameters. Model plays a fundamentalrole in this method, where the tuning result andcomputational requirement depend greatly on thestructure, accuracy and complexity of the model. Ifan operation around a given loading condition isconsidered, linear model may be enough to achievea good performance; otherwise, nonlinear analyticalor identified model must be employed due to thenonlinear behavior of the plant.

Besides the model, another feature that is con-sidered essential to this method is the objective func-tion. Any output or input variables of interest in thecontroller tuning can be included in the objective func-tion, and various types of functions can be designed.The following quadratic output performance objectivefunction over a finite horizon N is commonly used:

Jk =t=N∑t=1

(yk+t − r

)TQ

(yk+t − r

)(2)

where k is the current instant, yk+ t is the expectedmodel output vector at instant k+ t based on thecandidate controller parameters, which could includethe interested variables, such as power output, throttlepressure, and so on, r represents the correspondingreferences to these output variables, and Q is theweighting matrix.

If linear model is employed in the method, theminimization of the objective function can be for-mulated into a quadratic programming (QP) prob-lem, which is computationally efficient. However, inmost of the cases, nonlinear models are utilized to

better model the plant behavior; thus the optimiza-tion procedure becomes a non-convex problem withheavy computational load, and may get stuck in a localminimum far from the optimal value. Consequently,various heuristic search algorithms, such as GeneticAlgorithm (GA),14 Evolutionary Programming (EP),13

Particle Swarm Optimization (PSO),15 are proposed tofind the optimal controller parameters. A basic work-ing principle for these heuristic search techniques is:

1. Initialization: generate initial controller param-eter candidates randomly in the given solutionspace.

2. Evaluation: simulate the power plant modelwith these parameters and evaluate the corre-sponding objective function;

3. Modification: modify the parameters and con-tinue to evaluate the performance until findingthe satisfactory parameters.

The results in works13–15 show that, quality solu-tions and fast convergences can be provided in manyapplications. However, the nonlinear optimization isstill time consuming, which brings difficulties in mak-ing frequent online updates. Fortunately, the PI/PIDcontroller gains do not have to be updated for eachtime increment, thus a large window size can be cho-sen to tune the gains.15

Another drawback for this auto-tuning methodis the model-dependence character. An accuratemodel which can capture the dynamics of the powerplants over wide operation range is difficult todevelop. Therefore, it is worth mentioning here thatrecently, model-free direct tuning methods using theclosed-loop experimental data are employed in thepower plant to achieve an optimal tuning of the



PI/PID controller parameters. In Ref 11, iterativefeedback tuning (IFT) technique is employed for asimultaneously tuning of six main control loops inthe power plant. In Ref 16, extremum seeking (ES)technique is utilized to find the optimal PID gains ofthe superheater steam temperature control system inthe FFPP.

Similar to the ordinary model based methods,different overall control objective functions can alsobe designed in these approaches, but the gradientsearch based minimization of these objective functionsis time consuming and lacks robustness. One attractivefeature for these model-free based approaches is thatthey potentially eliminate or reduce the modelingeffort, but instead, a number of experiments arerequired for gradient computation, which have to bedesigned carefully to make sure the operation of theplant is not under too much stress.

The last technique introduced in this part isthe expert knowledge-based PI/PID auto-tuningapproach, where the expert knowledge is utilized todetermine and coordinate the controller gains. Forexample, to obtain a desired step response, a largecontrol signal is needed in the beginning to achieve afast rise time, which requires a big proportional gainand a small derivative gain. Also, because processeshave large time-delay, a small integral gain is desiredto reduce the overshoot. When the output response isnear the set-point, the proportional gain and integralgain should be changed from large to small and fromsmall to large, respectively, to make the controlledoutput converge to the set-point quickly.

On the above classical tuning knowledge, fuzzyrules are generated in Ref 17 to adjust the PIDparameters according to the current output errore(k) and its first difference value Δe(k), and a fuzzyauto-tuning PID controller is designed for the mainsteam pressure loop in the FFPP, which can enhancethe robustness and control performance of the PIDcontroller with fixed parameters.

The expert knowledge rule-table could be pro-duced off-line, and once finished; the PID parame-ters can be determined directly and efficiently by thelook-up table. However, no optimality can be achievedsince no optimization is performed in this approach.

Gain Scheduling PI/PID ControlThe computational complexity of the online-tuningof the PI/PID parameters leads to the developmentof gain-scheduling PI/PID control methodology, whichis more practical to deal with the limitation offixed parameter PI/PID controller and attain a betterwide-range operation performance in the power plant.

The essential idea of the gain scheduling controlis to change the controller parameters according to

the variations of process dynamics. A measurableprocess variable, which is descriptive of the operatingcondition, is known as a scheduling variable and usedto adjust the controller parameters.

For the FFPPs, the power load is usually selectedas the scheduling variable since its variation nat-urally represents various operating points of theplant, especially when the plant is operating under a‘constant-pressure’ mode. Then several typical loadingpoints are selected and at each point, the PID con-troller parameters are tuned offline. In online opera-tion, according to the value of the scheduling variable,the PID parameters can be determined through someinterpolations between the parameters predesignedat typical loading points.17–19 The block diagram ofa typical gain scheduling PID control is shown inFigure 12.

Robust ControlOne major issue in the control of FFPPs is theuncertainty. First, modeling mismatches are difficultto avoid due to the complex dynamics of the plantand the desire to use simplified model; and secondly,the unknown disturbances commonly exist all overthe plant due to the equipment wear, environmentalchanges, fuel variation, and so on.

In contrast to the adaptive approach whichattempts to learn the uncertainties of the plant andeventually adjusts the controller to be best suited forthe plant, the robust control has a fixed structurewhich yields acceptable performance for a given plantuncertainty set. Because the robust controllers aresimpler to implement without online-tuning to theplant variations, they have been extensively studied forFFPPs over the past few decades.

The H∞ control is the subject of the largest shareof the robust control researches in the FFPPs. Thebasic idea of the H∞ control comes from the theory inthe frequency domain. The H∞ norm of the transferfunction (the maximum singular value of the func-tion over the H∞ space) can be interpreted as a maxi-mum gain between bounded input energy and outputenergy; thus, developing a controller which could min-imize the H∞ norm as an objective function will natu-rally reduce the closed-loop impact of a perturbation,and enhance the stabilization or performance.20,21

To guarantee robust performance of the con-trollers under unexpected uncertainties, in Refs 22–25the H∞ control approaches are proposed to theboiler-turbine coordinated system, gas and oil-firedheating/cogeneration industrial boilers and gas tur-bine. In these works, an H∞ approach with mixedsensitivity has been utilized in the controller design,



Gain scheduler

PID

Controller

PID_1 PID_2 ... PID_n

FFPP

Controller parameters

Plant

output

Control

signal

ReferenceDriving

error

Scheduling

variable

FIGURE 12 | Gain Scheduling PID control.

K(s) G(s)

W1(s) W2(s) W3(s)

z1 z2 z3

e

d

u yr

–

FIGURE 13 | Block diagram of a typical H∞ mixed sensitivity problem.

which can achieve a trade-off between the robustnessand performance.

Figure 13 shows the block diagram of a typicalH∞ mixed sensitivity problem, where G(s) is thetransfer function of the plant, K(s) is the controller;the variables r, e, u, d, and y are the reference, controlerror, control input, disturbance input and systemoutput, respectively; W1(s), W2(s), and W3(s) areweighting matrices, and z1, z2, z3 are correspondingweighted control system outputs.

Performance and robustness are characterized byvarious well-known closed-loop functions, in partic-ular the sensitivity function S(s), the input sensitivityfunction R(s) and the complementary sensitivity func-tion T(s):

S (s) = (I + G (s)K (s))−1 , (3)

R (s) = K (s) (I + G (s)K (s))−1 = K (s) S (s) , (4)

T (s) = G (s)K (s) (I + G (s)K (s))−1 = I − S (s) , (5)

Here, ||S(s)||∞ reflects the disturbance rejectionand reference following ability of the system, ||R(s)||∞

reflects the allowed amplitude of additive uncertaintiesG(s)+Δ(s), ||T(s)||∞ reflects the allowed amplitude ofmultiplicative uncertainties (I+Δ(s))G(s).

A generalized plant G (s) can be built as follows:

⎡⎢⎢⎢⎣z1z2z3e

⎤⎥⎥⎥⎦ = G (s)[

ru

]=⎡⎢⎢⎢⎣W1 (s) −W1 (s)G (s)

0 W2 (s)0 W3 (s)G (s)I −G (s)

⎤⎥⎥⎥⎦[

ru

], (6)

u = K (s) e, (7)

and a transfer matrix from the disturbance input dto the control system outputs z= [z1, z2, z3]T can beconstructed as:

Φ (s) =⎡⎢⎢⎣

W1 (s) S (s)W2 (s)K (s) S (s)

W3 (s)T (s)

⎤⎥⎥⎦ . (8)

Then, shaping the closed loop transfer functionsS(s), R(s), and T(s) for the control system designis converted to find the controller K(s) which canminimize ||Φ(s)||∞.



In Ref 22, H∞ controller is compared with othermodern control design methods, such as 𝜇-synthesisand H2 approaches through the simulation on aboiler-turbine coordinated system model. Their testresults show that under the presence of complicatingfactors such as coupling between multi-variables, timedelay, modeling uncertainty, output disturbances andsensor noise in the FFPPs, the H∞ controller canprovide performance and robustness superior to thatattainable by other designs.

In Refs 26–28, the H∞ loop shaping techniqueis utilized to design multi-variable robust PI/PID con-trollers for the FFPPs. The results show that, owningto the advantages of the H∞ design approach suchas: yielding MIMO control laws, providing strongrobustness of the system, the proposed controllerscan achieve better performances than the conventionalPI/PIDs.

However, there are some drawbacks which maylimit the application of H∞ design approach inpractice:

1. Relying on the linear model of the plant;

2. The weighting matrices W1(s), W2(s), andW3(s) are important to attain a balance amongvarious control performance objectives, how-ever, the setting and tuning of them is indirectand inconvenient;

3. In general, the resulting controller has highorder, thus order reduction has to be carriedout without degradation in performance androbustness;

4. Cannot deal with the input constraints effec-tively (although tuning W2(s) can impose limi-tations on the inputs to some extent);

As the FFPPs are required to operate in a wideloading range, the liner model based H∞ approach isbecoming insufficient, even if the robustness can beguaranteed within a certain range around the typicaloperating point. Therefore, multi-model based H∞approaches are proposed recently.

In Ref 29, two transfer function modelsare developed at different operating point of theboiler-turbine unit of a coal-fired power plant, andtwo H∞ controllers are built based on these modelsto guarantee the robustness in each local-region.A bumpless switchover mechanism is designed toachieve a smooth transition between the two oper-ating regions, thus a wide range load following isattained. In Refs 30 and 31, robust H∞ trackingcontrollers are designed for the boiler-turbine unitvia Takagi-Sugeno (T-S) fuzzy model. Based on the

state-space type of local models, Lyapunov theoryand Linear Matrix Inequality (LMI) technique areemployed in these methods, and a robust trackingcontrol of the boiler-turbine is achieved in a wideoperation range with the stability of the closed-loopsystem being guaranteed.

Although H∞ approach is the most popularone in the area of FFPP robust control, it is byno means the only approach. Other robust controlapproaches such as linear quadratic Gaussian withloop transfer recovery32 (LQG/LTR) and 𝜇-synthesis33

are also applied to the FFPPs to enhance the systemrobustness and performance.

Model Predictive ControlAlthough better control performance can be attainedby the use of advanced PI/PID controller or robustcontroller, none of them can effectively deal with theinput constraints during the controller design stage.Therefore in practice, when physical limitations ofvalves in the FFPPs are involved, the performanceof the controllers will be degraded. In this context,the model predictive control (MPC) has been widelyemployed in recent years.

MPC refers to a class of control approacheswhich utilize an explicit process model to predict thefuture response of a plant and calculate the controlinputs through the minimization of a dynamic objec-tive function within the predictive horizon.34 Origi-nally applied mainly in the petrochemical industry, thepredictive control has progressed steadily and gradu-ally made a significant impact on the FFPPs control.The main reasons for its success in FFPP studies andapplications are:

1. It can effectively handle the actuator limitationsand allow the operation closer to constraints,which frequently leads to more rapid responseand more profitable operation.

2. It handles multivariable control problems natu-rally.

3. It can effectively deal with the large inertial andtime-delay behavior of the plant.

4. The tuning of the parameters is easy and intu-itive; no special attention needs to be placed onconstraints and optimization.

The basic idea and principle of the MPC isshown in Figure 14, and can be briefly explained inthree steps:

1. At current time k, based on the available infor-mation and predictive model, predict the futureresponse of the plant within the predictive



k k+1 k+Nuk+Ny

Past Future

Control horizon Nu

Predictive horizon Ny

yk+pyk–1

uk–1

rk + p

u*k+p

FIGURE 14 | Basic working principle of the MPC.

horizon Ny, and express the future predic-

tive outputs{

yk+1, yk+2, … , yk+Ny

}in terms of

future inputs{

uk+1,uk+2, … ,uk+Nu

}.

2. Through minimizing a given dynamic controlobjective function, calculate the optimal futurecontrol sequence

{u∗

k+1,u∗

k+2, … ,u∗

k+Nu

},

which can drive the predictive output y tofollow the reference r in an optimal way.

3. Send only the first input in the optimal sequenceto the plant, and repeat the entire calculation atsubsequent time.

Linear MPCThe first generation of MPCs applied to the FFPPs isthe linear MPCs, in which linear model for the plant isutilized and the optimal control inputs are determinedthrough minimizing a finite horizon quadratic perfor-mance objective:

J(k)=

Ny∑j=1

[y(k + j|k) − r

(k + j

)]T

× Q[y(k + j|k) − r

(k + j

)]+

Nu∑j=1

Δu(k + j − 1|k)T

RΔu(k + j − 1|k) .

(9)

The future projected output y can be relateddirectly back to the input vector Δu through the linearmodel and the objective function can be re-written inthe form of a standard quadratic programming (QP)problem with all input output constraints collected

into a matrix inequality involving the input vector. AsQP is one of the simplest possible optimization prob-lems, the optimal inputs can be found efficiently.34

Because the linear step response model of theplant can be easily obtained, dynamic matrix control(DMC) has been extensively studied. In Ref 35, DMCis applied to a drum-type boiler-turbine system toachieve a simultaneous control of electrical power,drum pressure and water level. It shows that thestep-response model based on the test data is bettersuited than the linearized model. In Refs 36 and37, DMC techniques have been applied to controlthe superheater/reheater steam temperature of theFFPPs, the results demonstrate that better controlperformance can be achieved as compared to theconventional PID control.

Another linear MPC that has been widely usedin the FFPPs is the generalized predictive control(GPC), where auto regressive moving-average modelis employed.38–41 In Refs 38 and 39, multi-variableGPCs are developed for a coordinated control ofthe boiler-turbine; their simulation results indicatethat, compared with PID, the GPCs can not onlytrack the target value smoothly and rapidly withsmaller overshoot and shorter adjusting time, butalso has stronger robustness. In Ref 40, GPC isdesigned to regulate the superheater and reheatersteam temperatures of a 200 MW power plant. Tofurther enhance the robustness of the system andattain a wide range operation of the plant, recursiveleast square based adaptive algorithm is added to theGPC, which can tune the controller parameters onlineusing the real-time input/output data. In Ref 41, theadaptive GPC is tested in a real power plant andevident improvements were observed in the results.

Nonlinear MPCAlthough the linear MPC can effectively improve thecontrol performance of FFPPs, due to the fact thatthe linear model works only for linear system, itsapplication is limited to a small operating region of aplant. For this reason, nonlinear plant model has beenused instead of the linear model in the MPC design toachieve a wide range plant operation.

To overcome the issues associated with nonlinearmodeling and computational requirement, artificialintelligence techniques have been applied. In Ref 42,the nonlinear analytical model of the plant is employedas the predictive model, and GA is utilized to cal-culate the optimal control sequence under the inputconstraints. In Ref 43, neural network-auto regres-sive exogenous (NN-ARX) model is identified for a200 MW oil-fired drum-boiler plant, and an MPC isdesigned based on the model. Simulation results show



that a satisfactory control of main steam pressureand temperature and reheat steam temperature canbe attained during load-cycling and other severe plantoperating conditions. To improve the computationalperformance of the nonlinear optimization, the par-ticle swarm optimization (PSO) and its modificationshave been used in Refs 44–46 to search for the opti-mal control sequence in the NN based nonlinear MPC;the advantages and effectiveness of these approacheshave been clearly shown through simulations of thesuperheater and reheater steam temperature control ofthe FFPPs. In Refs 15 and 47 online-update diagonalrecurrent neural network (DRNN) models have beendeveloped for 500 and 1000 MW FFPPs, and PSObased nonlinear MPCs are then designed to achievea plant-wide control.

Besides the intelligence-based nonlinear MPC,in Ref 48, a load-dependent exponential ARX(Exp-ARX) model that can effectively describethe plant nonlinear properties is identified off-line,and then used to establish a constrained multivariatemultistep predictive controller. Simulation studies ona 600 MW FFPP indicate that owing to the ability ofthe nonlinear model capturing the plant dynamics,much better control performance can be attainedwithout using the online-update method. Based onthe technique of input-output feedback linearization,GPC is combined with the nonlinear state feedbackcontroller in Ref 49 to solve the control problem of a160 MW boiler-turbine unit.

The nonlinear MPC is naturally suited for con-trol of the nonlinear FFPPs; however, there are twomain issues, which greatly limit its application: (1)the satisfactory nonlinear dynamic model is difficultto build; and (2) the nonlinear optimization lacks inrobustness and suffers from computational require-ment. To overcome these issues, the multi-model basedpredictive controllers are developed recently.

Multi-Model MPCThe essential idea of the multi-model technique is‘divide and conquer’, which uses a combination ofseveral linear models to approximate the nonlinearbehavior of the plant.50 Because the advances in linearmodeling and control theory can be directly takeninto account, the multi-model techniques bring analternative way to handle the nonlinearity, and itsintegration with the MPC approach has been shownto be effective to control the power plants.49,51–66

The earliest multi-model MPC used in powerplant control is presented in Ref 51, where networks ofdynamic local linear models are created after dividingthe whole operating region into a number of zones andthe global model is built by using the interpolation

among these local models. GPC is then designed onthe basis of the global model to achieve a wide rangecontrol of the power plant, as shown in Figure 15.Based on the Radial Basis Function (RBF) neuralnetwork and Adaptive Neuro-Fuzzy Inference System(ANFIS) approaches, local linear MPCs developedat different loading points are combined for thecoordinated control of a 500 MW plant.52,53 Althoughthe multi-model strategy seems more complicated thanthe direct nonlinear approaches, in fact, it is easierand more efficient to implement. To improve thewide-range operation of a boiler-turbine system, afuzzy interpolated DMC is proposed in Ref 54, whichcan be viewed as an extension of the linear DMC inRef 35.

In Refs 49, 55, 56, GPCs are developed onthe basis of neuro-fuzzy networks (NFNs), both thelocal linear model weighted and controller weightedapproaches are given in these works. Through thesimulation studies on the boiler-turbine coordinatedsystem and superheater steam temperature system,better control performance is shown compared to thelinear GPC.

In Refs 57–66, multi-model MPCs are pro-posed on the piecewise linear (PWL) models, piecewiseaffine (PWA) models or Takagi-Sugeno (T-S) fuzzymodel using various kinds of objective functions andcomputational tools, such as QP,57,58,60 LMIs,63–66

multi-parameter programming (MPT),59 GA61 anditerative learning control (ILC).62 Besides Ref 62, it isinteresting to note that, a state-space type of local lin-ear model is adopted in most of the multi-model MPCsbecause of the advances in multi-variable systemsand the-state-of-the-art control techniques for linearsystems.

In Refs 59 and 61, the stability of the closed-loopcontrol system is achieved and the overall controlperformance is improved by including a terminalinequality constraint, which forces the states at theend of the finite prediction horizon to lie within aprescribed terminal region, and by adding a quadraticterminal state penalty cost in the objective function.In Refs 63–66, an infinite-horizon objective functionis adopted and a Lyapunov function is constructed tofind the upper-bound of the objective function andensure the stability. The control input can be solvedby minimizing this upper-bound while subjecting tothe stability and input constraints in the form oflinear matrix inequalities (LMIs). In Refs 61, 65, 66, adisturbance observer and steady-state target calculator(SSTC) are added to the MPC structure to achievea tracking control of the boiler-turbine unit even inthe case of significant unknown disturbances or plantparameter variations.



Multi-model MPC

Moving horizon

optimizaiton

Global model

Local model #1

Local model #2

Local model #i

. . .

Weight

calculation

+

FFPPReference

Optimal control

input Plant output

Power load

FIGURE 15 | Control diagram ofmodel-weighted multi-model MPC.

Intelligent ControlThe development of intelligent techniques in powerplant control are increasing steadily in recent yearsowing to its advantages, such as overcoming the sig-nificant nonlinear and uncertain dynamics, computa-tional complexity and other problems associated withlarge-scale distributed complex power plants.

Intelligent control is a class of control tech-niques that use various computational intelligencetechniques to develop the computer-based control sys-tems. Because, the intelligent control system behavesas humans or species in nature, with the ability tolearn and discover knowledge, emulate human exper-tise and decision making, and accomplish the tasks, itcan autonomously achieve a high level goal even in thecomplicated or unexpected case.67

As was already introduced in the advancedPI/PID control and MPC sections, many differentintelligent techniques have been applied to powerplant control, and their structures are modified in gen-eral depending on the purpose of application. Themost popular intelligent techniques are neural net-work, fuzzy logic, evolutionary programming, geneticalgorithm, particle swarm optimization, multi-agentsystems, as well as their combinations.

The first intelligent technique investigated in thesection is the neural network (NN). Since the NNhas outstanding abilities in knowledge discovery, ordiscovering underlying, hidden patterns in data sets,it provides an effective way to develop the model forthe nonlinear power plant using only the input-outputdata, which is generally the first and foremost impor-tant step in the advanced controller design.

To provide dynamic information about the plantfor controller design, the usual NN modeling iscombining the conventional feedforward networks

(such as BP or RBF neural networks) with tappeddelays.68 Beside this, in Refs 44–46, Elman neuralnetworks (ENN) and their modifications are utilizedto approximate the behavior of the steam temper-ature systems of FFPPs. The ENN differs from theconventional feedforward networks in that it includesrecurrent or feedback connections. The delay in theseconnections store values in the previous time-step anduse them as inputs in the current step, which makes thenetwork sensitive to the history of input and outputdata, suitable for dynamic system modeling.

The basic structure of an ENN with M inputsand N outputs is shown in Figure 16, where therecurrent neurons are in the context layer. The outputsin each layer of an ENN are given by:

xj

(k)= f

(M∑

i=1

W1i,jui

(k)+

R∑i=1

W3i,jci

(k))

, (10)

ci

(k)= xi

(k − 1

), (11)

yj

(k)= g

(R∑

i=1

W2i,jxi

(k))

, (12)

where, W1i,j is the weight that connects node i in the

input layer to node j in the hidden layer; W2i,j is the

weight that connects node i in the hidden layer to nodej in the output layer; W3

i,j is the weight that connectsnode i in the context layer to node j in the hidden layer;and f (·) and g(·) are the transfer functions of the hiddenlayer and the output layer neurons, respectively, wheref (·) mostly takes logsig or tansig function and g(·) oftentakes purelin function.



…

…

… …

Δ

y1(k)

x1(k)

u1(k)C1(k)uM(k)

Cp(k)

yN(k)

xR(k)

Context layer

Δ

FIGURE 16 | Basic structure of the ENN.

As a special case of the ENN, the diagonal recur-rent neural network (DRNN) is also popular in thearea of FFPP dynamic modeling.13,15,47,69 Comparedwith the ENN, the context layer of DRNN is col-lapsed to the hidden layer, which eliminates the crosstalks and reduces the number of weights between thecontext layer and the hidden layer. Therefore, fewertraining iterations are needed and the DRNN is bettersuited for real-time application.69

In order to capture the complex dynamic of thewhole power plant, the DRNN based combined modelis proposed in Refs 15 and 69, where each primarycomponent of the plant is approximated by a DRNNmodel. Then these DRNNs are connected by using theoutputs of a DRNN as the inputs of another DRNN.The composed hierarchical structure can achieve asatisfactory accuracy, thus can be used as the plantsimulator.

With the dynamic process model being suc-cessfully developed, various advanced controllercan be designed, for example, the aforementionedAuto-tuning PI/PID Control and nonlinear MPC.To generate solutions to the highly complex nonlin-ear optimization problem, heuristic optimizationtechniques, such as evolutionary programming(EP),13,70 genetic algorithm (GA)14,42 or particleswarm optimization (PSO)15,44–47 have been employedextensively, providing quality solutions and fast con-vergence compared with the conventional nonlinearprogramming.

The basic principle for all these heuristic opti-mization algorithms can be concluded as: initializa-tion, evaluation and modification as we introduced inthe Auto-tuning PI/PID Control section. In Ref 71,basic GA and PSO algorithms are compared and theresults show that PSO can attain a better performance.However, because the search procedure by the PSOstrongly depends on the agent best value so far (pbest)and the global best value (gbest), the search area can belimited by them, which may lead to the local minima.

PlantNNIC

ΔΔ

yref

u(k) y(k+1)

FIGURE 17 | Control system of the NNIC.

Therefore, by introducing a natural selection mecha-nism, which is usually performed by the EP and GA,the effect of pbest and gbest is gradually curtailed anda broader search area can be realized. The resultinghybrid PSO can have a higher performance.

Besides the traditional model based controller,the neural network technique can also be used todesign the model-free controllers directly, which iscalled the neural network inverse control (NNIC).

The essential idea of the NNIC is to identify aninput/output mapping:

u(k)= g

(y(k + 1

), y

(k), y

(k − 1

), … , y

(k − n

),

u(k − 1

),u

(k − 2

), … ,u

(k − m

)), (13)

which can be viewed as a nonlinear inverse mappingof the conventional dynamic model:

y(k + 1

)= g

(y(k), y

(k − 1

), … , y

(k − n

),u

(k),

u(k − 1

), … ,u

(k − m

)). (14)

The inverse mapping function g(·) can be iden-tified by using the ENN or DRNN, and it is capableto find the control input u(k) to drive the system to anexpected point y(k+ 1). Therefore, it can be used asa controller by replacing the future expected outputy(k+1) in Equation (13) with the desired output, yref.If the network represents the exact inverse, the controlinput produced by the network will drive the systemoutput y(k+ 1) to yref. Figure 17 illustrates how theNNIC is applied as a controller in the system.

As the NNIC is a feedforward controller, in orderto deal with the unavoidable modeling mismatchesand the unknown plant variations and disturbances,a PID compensator is usually augmented to the sys-tem, and the resulting control structure is shownin Figure 18. The NNIC is used as a feedforwardcontroller, which provides the main contribution tothe control signal to make the plant respond quicklyto the set-point changes, thus shorten the stabilizationtime and reduce the overshot of the control process,and the PID compensator is serving as the feedbackcontroller to eliminate the steady-state control errorinduced by the NNIC.

The feedforward-feedback control system isutilized in Refs 72 and 73 to regulate the superheater



Plant

PID

compensator

NNIC

ΔΔ

+

yref

u(k) y(k+1)

FIGURE 18 | NNIC augmented by a PID compensator.

steam temperature in the FFPPs. The results in theseworks show that owing to the advantages of bothNNIC and the PID compensator, better control per-formance can be realized than the original cascadePID control.

Another intelligent technique being widely usedin the area of FFPP modeling and control is the fuzzylogic control (FLC), which has advantage in capturingthe tacit knowledge of the system. The Takagi-Sugeno(T-S) fuzzy modeling technique is now the most pop-ular FFPP modeling strategy.74 Its basic idea is usingthe fuzzy set to divide the whole operation regionof the system into several overlapping local regions;in each local region, a simple linear model is devel-oped to represent the local dynamics of the system,and then the global model is finally derived from theweighted summation of all the local models. Becausethe T-S fuzzy model can approximate the nonlinearbehavior of the plant using the combination of severallinear models, the advanced linear control theory canbe used to overcome the nonlinear control problem.Moreover, since the overlaps between different regionscan guarantee smooth transitions between local mod-els, the T-S fuzzy model naturally provides a bumplessswitchover.

Take the discrete-time, state-space form of localmodel, for instance, the T-S fuzzy model can bedescribed as follows:

Ri ∶ IF z1

(k)

is Mi1, z2

(k)

is Mi2, … , zN

(k)

is MiN , THEN ∶{

x(k + 1

)= Aix

(k)+ Biu

(k)

y(k)= Cix

(k)+ Diu

(k)

i = 1,2, … L

(15)

where Ri denotes the ith fuzzy rule, L the num-ber of fuzzy rules, Mi

j (j = 1,2, … ,N) the fuzzy sets;x(k)∈ℜn the state vector, u(k)∈ℜm the controlinput vector, and y(k)∈ℜl the output vector. Thematrices (Ai, Bi, Ci, Di) are local system matrices, andZ(k)= (z1(k), z2(k), … , zN(k)) is the antecedent vector

of the fuzzy model, which is composed by current andpast measurable variables of the plant.

By using fuzzy blending, the dynamic fuzzymodel (15) can be expressed by the following globalmodel: {

x(k + 1

)= AZx

(k)+ BZu

(k)

y(k)= CZx

(k)+ DZu

(k) (16)

where AZ =∑L

i=1 wi

(Z(k))

Ai, wi

(Z(k))

=Mi

(Z(k))

∕∑L

i=1 Mi(Z(k))

, Mi(Z(k))

=N∏

j=1Mi

j

(zj

(k))

, and all other matrices Bz, Cz, and

Dz are defined in the same way.In Refs 30, 60–62, 64–66 an approximation or

transformation of the nonlinear system has been usedto obtain the linear state-space model at different load-ing conditions, then according to the expert knowl-edge or the nonlinear investigation of the plant, thefuzzy rules are developed to connect the local mod-els and form the integrated T-S fuzzy model of theFFPP. In Ref 75, a data-driven fuzzy modeling strategyis proposed, where Gaustafson-Kessel (G-K) cluster-ing is used to provide an appropriate division of theoperation region and develop the structure of the T-Sfuzzy model. Then by combining the input data withthe corresponding fuzzy membership functions, thesubspace identification (SID) method is extended toextract the local state-space model parameters. Owingto the advantages of the both methods, the resultingfuzzy model can represent the boiler-turbine unit ofan FFPP very closely for advanced controller design.

The fuzzy technique also provides an effectiveway to design the controller. In Refs 76 and 77, controlrules are first given based on the expert knowledge todetermine the linguistic value of the control signal cor-responding to the current output error e(k) and its firstdifference valueΔe(k) . Fuzzy logic controllers are thendeveloped based on Gaussian and Triangular-shapedmembership functions to control the nonlinearboiler system and boiler-turbine system in the FFPP,respectively.

In Ref 78, a Fuzzy Auto-Regressive Mov-ing Average (FARMA) controller is applied to theboiler–turbine system, the controller is an inversemapping of the FARMA model, which finds thecontrol input according to the reference value andhistoric input–output data. Unlike a conventionalFLC, where an expert gives the linguistic values of theantecedent and consequent variables and makes rules;in the FARMA controller, these linguistic values aredetermined from the crisp values of the input–outputhistory at each sampling time. It has been shown



in the simulation that by using the self-organizingprocedure to update the rule base, better controls canbe applied as time progresses.

Meeting the power demand of the grid hasbeen the primary FFPP operation task. However, dueto the energy shortage and environmental concernworldwide as well as competition among utilities andother society driven forces, more stringent require-ments, such as minimization of fuel consumption,maximization of duty life and minimization of pollu-tion, etc., have to be fulfilled by the power plant toachieve an optimal operation. Therefore, not only theconventional dynamic control performance, but alsothe set-point optimization considering the aforemen-tioned multiple objectives is important to realize anintegrated optimal control of the FFPP, where vari-ous computational intelligence techniques are utilizedto overcome the nonlinear modeling and calculationissues of the FFPP.

In Refs 79–81, multi-objective optimization isperformed to adjust the power-pressure mappingto optimally accommodate different operating sce-narios, where more economic, environmental andsafety concerns are taken into account. Steady-stateANN models are first developed in these works tocapture the steady-state nonlinear behavior betweenmulti-variables in the boiler-turbine system, and thenPSO algorithm is employed to calculate the optimalset-points. In Refs 13, 15, 71, the similar methods areused in large-scale FFPP simulators to achieve a plantwide optimization of the set-points.

To reduce the nitrogen oxides (NOx) emissionduring coal combustion in the FFPPs, combustionoptimization is studied in Refs 82–85 to find the opti-mal openings for the primary, secondary and over-fireair valves at different layers of the boiler. Since thedevelopment of an accurate analytical model for acoal-fired boiler is difficult owing to the complexityof the system, data-driven modeling methods such asANN, support vector machine (SVM), fuzzy c-meansclustering as well as their mixtures are proposed tobuild the steady-state relationships between these airvalves and NOx emission.

The electric utility industry is charged to deliverpower as inexpensively and as reliably as possible.Meeting these dual obligations has become increas-ingly difficult over the past 30 years. Environmentaland economic concerns pressed the utility industryto develop clean and efficient ways of burning coaland oil. This has required major improvements innot only the optimization and control, but also themonitoring of electric power plant components suchas boilers. It has become a challenge to measure hightemperature distributions of high-pressure liquids,

steam, combustion gases, and heat transfer compo-nents in extremely adverse power plant environments.Traditional sensors have not exhibited sufficientstability and long-term accuracy without requiringexpensive maintenance and recalibration. However,intelligent distributed parameter estimation coupledwith the fiber-optic sensor system promises betterestimate of the temperature distribution of a boilerfurnace and for improved combustion.85,86 Thebasic approach in developing the intelligent mon-itoring system is in two folds: (1) development ofdistributed parameter system (DPS) models to mapthe three-dimensional (3D) temperature distributionfor the furnace; and (2) development of an intelligentmonitoring system for real-time monitoring of the3D boiler temperature distribution based on the 1Dfiber-optic sensors.

The aforementioned intelligent techniques havebeen applied extensively in the area of power plantmodeling, control, optimization and monitoring, andtheir effectiveness in dealing with the nonlinearityand uncertainty of the plants has been demonstratedthrough simulations. However, both the FFPP and itsoperation system are large-scale complex system con-sisting of many subsystems and functions, and it is dif-ficult and dangerous to manage the system using onlythe centralized control schemes or loosely decentral-ized control schemes because a single failure can bringdown the entire system. Therefore, recently, there hasbeen growing interest in multi-agent systems (MASs)in order to deal successfully with the complexity anddistributed problems in power plants and make thecontrol system operate at a higher level of automation,flexibility and robustness.

The basic unit in the MAS is the agent whichhas intelligent and autonomous properties becauseit is reactive, proactive, social, flexible and robust,and has multiple algorithms for solving problems.After receiving and confirming the objective throughthe communication with other agents, the agent willchoose a plan for the objective and select an algorithmto launch the plan.

The agents are arranged into a hierarchicalstructure and form the MAS. In the highest level ofthe MAS, the task delegation agent allocates tasksto different agents and investigates the performanceand problems of the lower level. In the middle level,the mediate agent coordinates the cooperation of thelower agents such as, asks a certain agent to share theinformation directly with the information requesters.A monitoring agent is also placed in this layer toselect, store and analyze the sensing data of the powerplant. In the lowest level, intelligent agents that havemultiple algorithm modules are cooperating together



to identify the system, perform the multi-objectiveoptimization and determine the set-points and controllaw. Design of single agents and the integrated MASas well as its implementation in the FFPPs are wellintroduced in Refs 13, 67, 79, 80.

ADVANCED CONTROL TECHNOLOGYPRODUCTS FOR THE FFPP

Although the power plant industry may still be reluc-tant to accept changes that would revolutionize wellassessed procedures due to the reasons such as: relia-bility, complex design, tuning parameters and train-ing of plant personnel, the environment of growingenergy and pollution issues, increased FFPP controlchallenges as well as the significant improvementsreported by the academic studies has already led tothe technology development in the industry. Moreover,over the course of the last two decades, the widespreaduse of distributed control system (DCS) in powerplant and the development of computer technologieshave facilitated the employment of advanced con-trol approaches. The new control software could beinstalled in the programmable logic controllers (PLCs)or industrial computers and communicates with theDCS through the object linking and embedding forprocess control (OPC) or Modbus protocols, whichwill then collect the required data and send the com-mands to regulate the actuators. For these reasons,several companies have presented their own commer-cial FFPP control solution products, aiming at improv-ing the operation performance of the FFPPs. Thepurpose of this section is to introduce several examplesof commercial products provided by different compa-nies to demonstrate the effectiveness of the advancetechnologies in the FFPP practice.

The first product introduced is the BCOS-ZOLOoffered by ZOLOBOSS,87 which is an FFPP opera-tion optimizer on the basis of direct monitoring ofthe combustion zone in the furnace and combustionoptimization. As is well known, the combustion inthe furnace is the energy source of the FFPP, how-ever, the high temperature (beyond 1500∘C) in thecombustion zone of the furnace makes it impossi-ble to measure the combustion condition (tempera-ture, and the contents of O2, CO, etc., in the fluegas) using the traditional instrumentations such as thethermocouple, Zirconium oxide detector, CO detec-tor, and so on. Therefore, so far, the plant personnelcan only estimate the parameters in the combustionzone according to the experience; based on the mea-surement data around the rear flue gas channel, whichis inaccurate and thus brings great challenges in the

combustion control and the resulting difficulties insteam temperature and pressure control.

For this reason, a novel laser based measure-ment product is developed by the ZOLOBOSS basedon the technology of tunable diode laser absorptionspectroscopy (TDLAS). The TDLAS instruments relyon well-known spectroscopic principles and sensitivedetection techniques, coupled with advanced diodelasers and optical fibers developed by the telecommu-nications industry. The principles are straightforward:Gas molecules absorb energy at specific wavelengths inthe electromagnetic spectrum. At wavelengths slightlydifferent from these absorption lines, there is essen-tially no absorption. Thus, by (1) transmitting a lightbeam through the gas mixture in the boiler furnacewhere the target gas is contained, and (2) tuning thewavelength of the beam to one of the absorption linesof the target gas, and (3) accurately measuring theabsorption of that beam, one can deduce the concen-tration of target gas molecules as well as the flue gastemperatures distributed over the beam’s path. In gen-eral, multiple light paths are assigned in the grid formon the certain layer of the furnace, so that throughthe computer aided calculation and tomography, thetemperature and gas concentration distribution can beobtained and viewed.

Knowing the combustion condition in the fur-nace will provide a direct guidance for the FFPP oper-ators to achieve a safe and efficient operation of theboiler; for example, prevent the dust deposition andslagging in the furnace, reduce the deviation of theflue gas temperature at the exit of the furnace to pre-vent the explosion of the heating surface, etc. More-over, by using the data in the combustion zone ofthe furnace, more accurate combustion model can bedeveloped or updated. The BCOS-ZOLO proposes theuse of the ANN to capture the dynamics between theboiler operation condition (load, fuel variations, envi-ronmental temperature, etc.), distributions of fuel andair, and the performance of the combustion (boilerefficiency, NOx emission, etc.). On the basis of themodel, the best distributions of the fuel and aircan be calculated for different loading condition andfuel variation through multiple computational intel-ligence algorithms to realize an optimal operation ofthe boiler.

Furthermore, the flue gas temperature in the fur-nace can be used to calculate the equivalent radiationenergy in the boiler, which can represent a balancebetween the boiler and turbine. Thus, by using theradiation energy signal as a feedforward signal, theissues of large inertia behavior of the steam pressureand temperature control system can be overcome to a



large extent, and a better load following operation candirectly be achieved.

A similar TDLAS based FFPP optimization isalso offered by the SIEMENS SPPA-P3000 Processoptimizer.88 Both the two products have now beenapplied in more than 20 FFPPs in the world, mostlyin the United States, Germany and Korea. The mainfeatures of them can be concluded as:

1. Direct measuring of the parameters in the com-bustion zone of the boiler through the TDLAS;

2. Developing the combustion model using theANN technique;

3. Using the computational intelligence technolo-gies to optimize the combustion;

4. Improving the control performance of the steampressure and temperature.

To improve the control performance ofthe large-scale power plant, many companiessuch as SIEMENS,89 EMERSON,90 ABB91 andHONEYWELL,92 have launched their own FFPPcoordinated control optimizers. The common featuresof these commercial products are: (1) identifyingthe process model to approximate the dynamics ofthe plant; and (2) utilizing advanced control strate-gies such as multi-variable control methods, MPCsor intelligent control algorithms to improve theperformance of the plant.

Among these commercial products, the PROFIUCC offered by SIEMENS89 is the most widely usedone. The essential idea of the PROFI UCC is devel-oping/updating a nonlinear model of the boiler andusing the technology of MPC to estimate the ‘ther-mal energy’ produced by the boiler, through which,a pre-action command can be given to the fuel feederto compensate for the inertia and delay behavior ofthe boiler so that a quick load following performanceand a smooth response of the main steam pressure andtemperature can be achieved.

On the basis of model prediction, a multi-inputmulti-output rate-optimal controller (MIMOROC)is proposed by the HONEYWELL’s UES92 for theoptimization of the combustion, steam pressure andtemperature control. By using the MPC’s abilityof dealing with the input–output constraints, theMIMOROC allows operation closer to the constraintscompared with conventional control, which leads toquicker load following and more efficient and prof-itable operation. Remarkable plant efficiency andprofit improvement have been reported on the FFPPs,where these advanced plant operation optimizers areimplemented.

DISCUSSIONS AND CONCLUSIONS

Over the past hundred years, FFPPs have been theprimary power generation plants in the world, andmade a solid foundation for the people’s lives, andsocial and industrial developments. Design of FFPPhas evolved dramatically in the area of scale, workingparameters, configurations and techniques to meet thegrowing power demand.

In recent 20 years, in order to solve the controlissues of FFPPs such as wide range load following,fuel and plant behavior variations, and achieve a moresafe and efficient operation, various advanced FFPPcontrol techniques have progressed swiftly in boththe academic researches and industrial applications,challenging the classical PI/PID based control system.The major FFPP control developments include:

1. Advanced PI/PID controllers using auto tuningor gain scheduling techniques to improve theoperation of the FFPPs in a wide load range;

2. Robust controllers in order to deal with the fueland plant behavior variations, uncertainties anddisturbances;

3. Linear, nonlinear or multi-model based MPCs tohandle the large-inertia behavior and the strictinput–output constraints of the plants;

4. Computational intelligence techniques in model-ing, optimization and control, solving the non-linear issues of the plants.

These advanced control methods as well as theirmixtures have all been implemented in the FFPPcontrol practices and we cannot answer which oneis the best and will be the future trends of theFFPP control. But, considering the benefits brought bythese advanced technologies and the facilities providedby the wide use of DCS and fast development ofcomputer technologies, it is safe to say that much moredevelopment in nonlinear, intelligent and model-basedMIMO control of FFPPs is ahead of us, and theadvanced control techniques will replace the classicalPI/PID controllers in a foreseeable future.

However, through the investigation of manyadvanced FFPP control works, there are several issuesthat may need to be studied further in the future:

1. Measurement of key parameters during theplant operation. For example, novel mea-surement technologies can be developed tomonitor the pulverized coal concentration andprimary air flow rate from pulverizing mills tothe burners (Some products have already been



developed recently, for example, the PfMasterby ABB/Greenbank). These parameters areunknown in the FFPPs, but have been found tobe the main factors causing the uneven distri-bution of the temperature field in the furnaceand the explosions of the waterwalls and super-heaters. Therefore, through the measurement ofthem, additional regulators can be installed inthe outlets of the pulverizing mills to guaranteeeven and balanced combustion in the furnace.

Another example is the coal properties. Cur-rently the variation of the fuel is one of thekey problems in the coal-fired power plants,the uncertainty greatly influences the safety,stability and efficiency of the combustion andplant operation. Therefore, soft measurementtechnologies of the coal properties are expectedto be developed using the combination of ther-modynamic knowledge and plant operationdata, so that more optimal and robust controlcan be attained through its estimation.

2. Optimal control. As mentioned before, in thecontext of global energy and environmen-tal issues, the control objective of the FFPPshould be broadened from a simple objectiveof dynamic control optimality to simultane-ous and multiple objectives of plant economicoperation, emission and dynamic control perfor-mance. Thus, on the one hand, advanced multi-objective optimization with Pareto optimalitycan be developed based on the plant models anddynamics and environmental changes to provideoptimal set-points for the control layer, replac-ing current fixed and non-optimal set-pointsgiven at specific loading condition; on the otherhand, more simple and efficient method suchas novel economic MPC can be investigated,where both the dynamic control objective andother non-quadratic economic objectives areintegrated together in the controller.

3. Closed-loop data-driven modeling. Modelingis the first and foremost important step inadvanced controller design. In most of thestudies, the researchers focused on the controlalgorithm, but failed to consider the modeldevelopment. In many of their works, theresults are relying on the availability of thenonlinear analytical model. However, con-sidering the complexity of real FFPP, it isdifficult to develop an accurate analytical modelwithout the knowledge of thermodynamics

and design specifications of many components,which becomes the major limitation for theapplication of advanced control methods. Forthis reason, system identification methods usingthe closed-loop plant operation data providean efficient way to develop a control orientedmodel in practice. In that case, how to design anoptimal input signal for identification, examineand polish the data, select the model structure,and test the model sufficiently for healthy inter-action with the identification theory, are allimportant issues in the FFPP practices.

4. Better combination of the analytical model andmodern control theory. Although difficult todevelop, an analytical model is after all thebest one to accurately portray the dynamicsof nonlinear FFPPs; it can also provide muchinformation which will guide the controllerdesign. Therefore, how to better integrate theknowledge of analytical model and advancedcontrol technology to develop a controller bettersuited for the FFPP is a possible future researchtopic, for example, leading to an analyticalmodel based nonlinear predictive controller.

5. Reliability. Although the advance controllerswill effectively deal with the control issues ofthe FFPP and greatly improve the operationalperformance, we have to admit that, comparedto the single-input, single-output PI/PID controlloops currently used in FFPPs, these advancedcontrollers are complicated in structure, param-eter tuning and computation; therefore, it haslimitations in reliability. Furthermore, becausethe multi-variable controller would work in acentralized control unit in most of the applica-tions, a failure of the control algorithm wouldinduce a failure of the entire plant. These arealso the main reasons that the plant person-nel are reluctant to employ the modern controltechniques.

Besides careful design and testing of the con-troller and training of the plant personnel, apossible way to directly ensure the reliabilityof the system is keeping the well-assessed SISOPID loops present in the control structure. Thus,the new control system could be disconnected atanytime, without compromising the plant safety.In this case, study of bumpless switchover mech-anism between advanced controller and PIDcontroller could be made.



ACKNOWLEDGMENTS

This work was supported in parts by the National Natural Science Foundation of China (NSFC) underGrant 51036002, Grant 11190015 and Grant 51306082, the Doctoral Fund of the Ministry of Education ofChina under Grant 20130092110061, the Natural Science Foundation of Jiangsu Province, China under GrantBK20141119, the Cooperative Innovation Foundation of Jiangsu Province-Prospective Joint Research Projectunder Grant BY2013073-07 and the U.S. National Science Foundation under Grant ECCS 0801440.

REFERENCES1. Flynn D, ed. Thermal Power Plant Simulation and

Control. Stevenage, UK: IEE Press; 2003.

2. Souza GFM. Thermal Power Plant Performance Analy-sis. Heidelberg: Springer; 2012.

3. Gilman GF, Boiler control system engineering, ISA,Research Triangle Park, USA, 2005.

4. Raja AK, Srivastava AP, Dwivedi M. Power Plant Engi-neering. New Delhi: New Age International Publishers;2006.

5. Lindsey D. Power-Plant Control and Instrumentation:The Control of Boilers and HRSG Systems. Stevenage,UK: IEE Press; 2000.

6. Quazza G, Ferrari E. Role of power station control inoverall system operation. In: Proceedings of Symposiumon Real-Time Control of Electric Power Systems, Baden,Switzerland, 215–257, 1972.

7. Russell T. Utility front end controls. Instrument Societyof America, paper #88-0418, 113–122, 1988.

8. Landis R, Wulfsohn E. The control philosophy for a unitcontrol system for co-ordinated operation of a boilerand turbine. Electron 1988, February:19–23.

9. Gery HC. The evolution of coordinated control. Instru-ment Society of America, paper #88-0417, 109–112,1988.

10. Babcock & Wilcox. Chapter 41: Controls for fossilfuel-fired steam generating plants. In: Steam: Its Gener-ation and Use. 40th ed. New York: Babcock & Wilcox;1992.

11. Zhang S, Taft CW, Bentsman J, Hussey A, Petrus B.Simultaneous gains tuning in boiler/turbine PID-basedcontroller clusters using iterative feedback tuningmethodology. ISA Trans 2012, 51:609–621.

12. Garduno-Ramirez R, Lee KY. Compensation ofcontrol-loop interaction for power plant wide-rangeoperation. Control Eng Pract 2005, 13:1475–1487.

13. Heo JS, Lee KY. A multi-agent system-based intelligentheuristic optimal control system for a large-scale powerplant. In: Proceedings of the IEEE World Congress onComputational Intelligence, Vancouver, Canada, July16–21, 5693–5700, 2006.

14. Dimeo R, Lee KY. Boiler-turbine control system designusing a genetic algorithm. IEEE Trans Energy Conver1995, 10:752–759.

15. Lee KY, Van Sickel JH, Hoffman JA, Jung W-H, KimS-H. Controller design for a large-scale ultra supercritical once-through boiler power plant. IEEE TransEnergy Conver 2010, 25:1063–1070.

16. Fei W, Li Y, Shen J, Xiang X. Optimization ofsuperheated steam temperature control system usingextremum seeking algorithm. J Southeast Univ 2010,40:952–956.

17. Li S, Liu H, Cai WJ, Soh YC, Xie LH. A new coordinatedcontrol strategy for boiler-turbine system of coal-firedpower plant. IEEE Trans Contr Syst Technol 2005, 13:943–954.

18. Garduno-Ramirez R, Lee KY. Fuzzy gain-schedulingPID+decoupling control for power plant wide-rangeoperation. In: Proceedings of International Conferenceon Intelligent Systems Application to Power Systems,Kaohsing, Taiwan, November 5–8, 233–238, 2007.

19. Garduno-Ramirez R, Lee KY. Power plant fuzzy PIDscheduling control over full operating space. In: Pro-ceedings of the International Conference on Intel-ligent System Application to Power Systems (ISAP2003), CD ISAP03-086.pdf, Lemnos, Greece, August31–September 3, 2003.

20. Zames G. Feedback and optimal sensitivity: modelreference transformations, multiplicative seminorms,and approximate inverses. IEEE Trans Autom Control1981, 26:301–320.

21. Doyleand JC, Stein G. Multivariable feedback design:concepts for a classical/modern synthesis. IEEE TransAutom Control 1981, 26:4–16.

22. Zhao H, Li W, Taft C, Bentsman J. Robust controllerdesign for simultaneous control of throttle pressure andmegawatt output in a power plant unit. In: Proceedingsof the 1999 IEEE International Conference on ControlApplications, 802–807, Kohala Coast, HI August 1999.

23. Bentsman J, Zheng K, Taft CW. Advance boiler/turbinecontrol and its benchmarking in a coal-fired powerplant. In: Proceedings of the 14th Annual Joint ISAPOWID/EPRI Controls and Instrumentation Confer-ence, Colorado Springs, CO, June 2004.

24. Pellegrinetti G, Bentsman J. H∞ controller design forboilers. Int J Robust Nonlin Cont 1994, 4:645–671.

25. Najimi E, Ramezani MH. Robust control of speed andtemperature in a power plant gas turbine. ISA Trans2012, 51:304–308.



26. Nademiand H, Tahami F. Robust controller design forgoverning steam turbine power generators. In: Interna-tional Conference on Electrical Machines and Systems,Tokyo, Japan, November 2009.

27. Tan W, Niu Y, Liu J. H∞ control for a boiler-turbineunit. In: Proceedings of the 1999 IEEE InternationalConference on Control Applications, 910–914, KohalaCoast, HI August 1999.

28. Tan W, Marquez HJ, Chen T. Multivariable robust con-troller design for a boiler system. IEEE Trans ControlSyst Technol 2002, 10:735–742.

29. Zheng K, Bentsman J, Taft CW. Full operating rangerobust hybrid control of a coal-fired boiler/turbineunit. J Dyn Syst Meas Control 2008, 130:041011-1–041011-14.

30. Wu J, Nguang SK, Shen J, Liu G, Li Y. Robust H∞tracking control of boiler–turbine systems. ISA Trans2010, 49:369–375.

31. Liu S, Liu X, Shen J, Li Y, Wu J. Design of trackingcontroller for coordinated boiler-turbine control systembased on fuzzy Lyapunov functions. Proc CSEE 2013,33:96–103.

32. Park Y, Choi M, Lee J, Kim B, Lee KY. An auxiliaryLQG/LTR robust controller design for cogenerationplants. IEEE Trans Energy Conver 1995, 11:407–413.

33. Weng C, Ray A. Robust wide-range control ofsteam-electric power plants. IEEE Trans ControlSyst Technol 1997, 5:74–88.

34. Qin SJ, Badgwell TA. A survey of industrial modelpredictive control technology. Control Eng Pract 2003,11:733–764.

35. Moon U, Lee KY. Step-response model development fordynamic matrix control of a drum-type boiler-turbinesystem. IEEE Trans Energy Conver 2009, 24:423–430.

36. Sanchez-Lopez A, Arroyo-Figueroa G, Villavicencio-Ramirez A. Advanced control algorithms for steamtemperature regulation of thermal power plants. ElectrPower Energy Syst 2004, 26:779–785.

37. Kim W, Moon U, Lee KY, Jung W, Kim S.Once-through boiler steam temperature control usingDynamic Matrix Control technique. In: 2010 IEEEPES General Meeting, Minneapolis, MN, 2010.

38. Karampoorian HR, Mohseni R. Generalized modelpredictive control for a multivariable boiler-turbineunit. In: 11th International Conference on Control,Automation and Systems, Kintex, Korea, 811–814,2011.

39. Hou G, Xi Y, Liu J, Zhang J. Simulation research ofthe multi-variable generalized predictive control in 500MW Unit Plant Coordinated Control System. In: 2011International Conference on Advanced MechatronicSystems, Zhengzhou, China, 196–201, 2011.

40. Hogg BW, El-Rabaie NM. Multivariable generalizedpredictive control of a boiler system. IEEE Trans EnergConver 1991, 6:282–288.

41. Moelbak T. Advanced control of superheater steamtemperatures—an evaluation based on practical appli-cations. Control Eng Pract 1999, 7:1–10.

42. Wu J, Shen J, Krug M, Nguang SK, Li Y. GA-based non-linear predictive switching control for a boiler-turbinesystem. J Contr Theor Applicat 2012, 10:100–106.

43. Prasad G, Swidenbank E, Hogg BW. A neural netmodel-based multivariable long-range predictive con-trol strategy applied in thermal power plant control.IEEE Trans Energy Conver 1998, 13:176–182.

44. Ma L, Ge Y, Cao X. Superheated steam temperaturecontrol based on improved recurrent neural networkand simplified PSO algorithm. App Mech Mat 2012,128-129:1065–1069.

45. Lee KY, Ma L, Boo C, Jung W, Kim S. Intelligentmodified predictive optimal control of reheater steamtemperature in a large-scale boiler unit. In: 2009 IEEEPower Engineering Society General Meeting, Calgary,Canada, 2009.

46. Ma L, Ge Y, Lee KY. An improved predictive optimalcontroller with elastic search space for STC of a LSSPU.In: 51th IEEE Conference on Decision and Control,Maui, TX, 7024–7029, 2012.

47. Lee KY, Heo JS, Hoffman JA, Kim S, Jung W. Modifiedpredictive optimal control using neural network-basedcombined model for large-scale power plants. In: 2007IEEE PES General Meeting, Tampa, FL, 2007.

48. Peng H, Ozaki T, Haggan-Ozaki V, Toyoda Y. Anonlinear exponential ARX model-based multivari-able generalized predictive control strategy for thermalpower plants. IEEE Trans Control Syst Technol 2002,10:256–262.

49. Liu X, Guan P, Chan CW. Nonlinear multivariablepower plant coordinate control by constrained pre-dictive scheme. IEEE Trans Contr Sys Technol 2010,18:1116–1125.

50. Murray-Smith R, Johansen TA. Multiple ModelApproaches to Modeling and Control. London, UK:Taylor & Francis; 1997.

51. Prasad G, Swidenbank E, Hogg BW. A local model net-works based multivariable long-range predictive controlstrategy for thermal power plants. Automatica 1998,34:1185–1204.

52. Hou G, Liu H, Sun Y, Zhang J. Multi-model predic-tive function control based on neural network and itsapplication to the coordinated control system of powerplants. In: 2010 Chinese Control and Decision Confer-ence, Xuzhou, China, 3950–3954, 2010.

53. Wang D, Huang B, Meng L, Han P. Predictive con-trol for boiler-turbine unit using ANFIS. In: 2009 Inter-national Conference on Test and Measurement, HongKong, China, 2009.

54. Moon U, Lee KY. An adaptive dynamic matrix con-trol with fuzzy-interpolated step-response model for adrum-type boiler-turbine system. IEEE Trans EnergyConver 2011, 26:393–401.



55. Liu XJ, Chan CW. Neuro-fuzzy generalized predictivecontrol of boiler steam temperature. IEEE Trans EnergyConver 2006, 21:900–908.

56. Liu X, Liu J. Constrained power plant coordinated pre-dictive control using neurofuzzy model. Acta Automat-ica Sinica 2006, 32:785–790.

57. Novak J, Chalupa P. Predictive control of aboiler-turbine system. In: 16th WSEAS Interna-tional Conference on Circuits and Systems, Kos Island,Greece, 2012.

58. Hlava J, Hubka L, Tuma L. Modeling and predic-tive control of a nonlinear power plant reheater withswitched dynamics. In: 16th International Conferenceon Methods and Models in Automation and Robotics,Miedzyzdroje, Poland, 284–289, 2011.

59. Keshavarz M, Barkhordari Yazdi M, Jahed-MotlaghMR. Piecewise affine modeling and control of aboiler-turbine unit. Appl Therm Eng 2010, 30:781–791.

60. Wu K, Zhang T, Lv J, Xiang W. Model predictivecontrol for nonlinear boiler-turbine system based onfuzzy gain scheduling. In: Proceedings of the 2008 IEEEInternational Conference on Automation and Logistics,1115–1120, 2008.

61. Li Y, Shen J, Lee KY, Liu X. Offset-free fuzzy modelpredictive control of a boiler-turbine system based ongenetic algorithm. Simul Model Pract Theory 2012,26:77–95.

62. Liu XJ, Kong XB. Nonlinear fuzzy model predictiveiterative learning control for drum-type boiler-turbinesystem. J Process Control 2013, 23:1023–1040.

63. Wu X, Shen J, Li Y. Control of boiler-turbine coordi-nated system using multiple-model predictive approach.In: 2010 8th IEEE International Conference on Controland Automation (ICCA), Xiamen, China, June 9–11,2010.

64. Wu X, Shen J, Li Y, Lee KY. Stable model predic-tive control based on TS fuzzy model with applicationto boiler-turbine coordinated system. In: Proceedingsof the 50th IEEE Conference on Decision and Con-trol and European Control Conference (CDC-ECC),3356–3361, 2011.

65. Wu X, Shen J, Li Y, Lee KY. Stable model predictivetracking control for boiler-turbine coordinated controlsystem. IFAC Proc 2012, 8:201–206 8th Power Plantand Power System Control Symposium, PPPSC 2012.

66. Wu X, Shen J, Li Y, Lee KY. Hierarchical optimizationof boiler-turbine unit using fuzzy stable model predic-tive control. Control Eng Pract 2014, 30:112–123.

67. Lee KY. Intelligent techniques applied to power plantcontrol. In: 2006 IEEE Power Engineering SocietyGeneral Meeting, Montreal, Canada, 2006.

68. Ma L, Ge Y. Superheated steam temperature predictiveoptimal control based on external time-delay BP neuralnetwork and a simpler PSO algorithm. In: Proceedings

of the 31st Chinese Control Conference, Hefei, China,July 25–27, 2012.

69. Lee KY, Heo JS, Hoffman JA, Kim S, Jung W. Neuralnetwork-based modeling for a large-scale power plant.In: 2007 IEEE PES General Meeting, Tampa, FL, 2007.

70. Ghezelayagh H, Lee KY. Intelligent predictive controlof a power plant with evolutionary programming opti-mizer and neuro-fuzzy identifier. In: Proceedings of the2002 Congress on Evolutionary Computation 2002,2:1308–1313.

71. Heo JS, Lee KY, Garduno-Ramirez R. Multiobjectivecontrol of power plants using particle swarm opti-mization techniques. IEEE Trans Energy Conver 2006,21:552–561.

72. Ma L, Lin Y, Lee KY, Superheater steam tempera-ture control for a 300 MW boiler unit with inversedynamic process models. In: 2010 IEEE Power Engi-neering Society General Meeting, Minneapolis, MN,2010.

73. Lee KY, Ma L, Boo CJ, Jung WH, Kim SH. Inversedynamic neuro-controller for superheater steam temper-ature control of a large-scale ultra super critical (USC)boiler unit. In: Proceedings of the IFAC Symposiumon Power Plants and Power Systems Control, Tampere,Finland, July 5–8, 2009.

74. Takagi T, Sugeno M. Fuzzy identification of systems andits application to modeling and control. IEEE Trans SystMan Cybern 1985, 15:116–132.

75. Wu X, Shen J, Li Y, Lee KY. Data-driven modelingand predictive control for boiler-turbine unit usingfuzzy clustering and subspace methods. ISA Trans 2014,53:699–708.

76. Liu X, Chai T. Fuzzy logic strategy on boiler controlproblem. In: Proceedings of the 1997 American ControlConference, 1264–1265, Albuquerque, NM, June 44–6,1997.

77. Chang J, Lee KY, Garduno-Ramirez R. Multiagent con-trol system for a fossil-fuel power unit. In: 2003 IEEEPower Engineering Society General Meeting, Toronto,Canada, July 13–17, 2003.

78. Moon U, Lee KY. A boiler-turbine system con-trol using a fuzzy auto-regressive moving average(FARMA) model. IEEE Trans Energ Conver 2003, 18:142–148.

79. Head JD, Gomes JR, Williams CS, Lee KY. Implemen-tation of a multi-agent system for optimized multiob-jective power plant control. In: 2010 North AmericanPower Symposium, Arlington, TX, September 26–28,2010.

80. Lee KY, Head JD, Gomes JR, Williams CS. Multi-agentsystem based intelligent distributed control systemfor power plants, In: 2011 IEEE Power Engineer-ing Society General Meeting, Detroit, MI, July 24–28,2011.

81. Heo JS, Lee KY, Garduno-Ramirez R. Dynamic mul-tiobjective optimization of power plant using PSO



techniques. In: 2005 IEEE Power Engineering Soci-ety General Meeting, San Francisco, CA, July 12–17,2005.

82. Zhou H, Cen K, Fan J. Modeling and optimization ofthe NOx emission characteristics of a tangentially firedboiler with artificial neural networks. Energy 2004,29:167–183.

83. Lv Y, Liu J, Yang T. Nonlinear PLS integratedwith error-based LSSVM and its application toNOx modeling. Ind Eng Chem Res 2012, 51:16092–16100.

84. Lv Y, Liu J, Yang T, Zeng D. A novel least squaressupport vector machine ensemble model for NOx emis-sion prediction of a coal-fired boiler. Energy 2013,55:319–329.

85. Lee KY, Velas JP, Kim BH. Development of anintelligent monitoring system with high temperaturedistributed fiber-optic sensor for fossil fuel power plant.In: Proceedings of the IEEE Power Engineering SocietyGeneral Meeting, PESGM2004-001350.PDF, Denver,CO, June 6–10, 2004.

86. Kim BH, Velas JP, Lee KY. Development of intelligentmonitoring system for fossil-fuel power plants usingsystem type neural networks and semigroup theory. In:

Proceedings of the IEEE Power Engineering SocietyGeneral Meeting, San Francisco, CA, June 12–17, 2005.

87. ZOLO Technologies. Coal power plant combus-tion optimization. Available at: http://zolotech.com/power-generation/coal-power-plant-efficiency/.(Accessed December 2013).

88. SIEMENS Power Plant Automation. SPPA-P3000solutions for process optimization. Available at: http://www.energy.siemens.com/hq/en/automation/power-generation/sppa-p3000/. (Accessed February 2014).

89. KWU LPSC Technology Center. PROFI unit coordi-nated control system, Siemens AG, 2000.

90. EMERSON Process Management. Smart processoptimization solutions. Available at: http://www2.emersonprocess.com/en-us/brands/smartprocess/pages/index.aspx. (Accessed March 2014).

91. ABB. OPTIMAX, plant optimization solutions forpower generation. Available at: http://www05.abb.com/global/scot/scot267.nsf/veritydisplay/e15f7798384e6f02852573a3004c9a5a/$file/plant_optimization_s_deabb_1291_06_e.pdf. (Accessed May 2014).

92. Honeywell. Honeywell UES. Available at: https://www.honeywell.com. (Accessed March 2014).


http://zolotech.com/power-generation/coal-power-plant-efficiency/

http://www.energy.siemens.com/hq/en/automation/power-generation/sppa-p3000/

http://www2.emersonprocess.com/en-us/brands/smartprocess/pages/index.aspx

http://www05.abb.com/global/scot/scot267.nsf/veritydisplay/e15f7798384e6f02852573a3004c9a5a/$file/plant&uscore;optimization&uscore;s&uscore;deabb&uscore;1291&uscore;06&uscore;e.pdf

https://www.honeywell.com

steam power plant configuration, design, and...

Documents