Vous êtes sur la page 1sur 12

Control Eng. Practice, Vol. 5, No. 3, pp.

301-312, 1997

Pergamon PII:S0967-0661 (97)00007-5

Copyright 1997 Elsevier Science Ltd Printed in Great Britain. All rights reserved 0967-0661/97 $17.00 + 0.00

BOILER-TURBINE DYNAMICS IN POWER-PLANT CONTROL


C. Maffezzoni

Dip. Elettronica e lnformazione, Politecnico di Milano, P.zza L. da Vinci, 32, Milano, Italy (maffezzo@elet.polimi.it)

(Received October 1996; in final form January 1997)

Abstract: Fossil-fired power plants are generally equipped with complex control systems, the structure of which reflects the relevant process dynamics underlying the steamgeneration process. This paper highlights the principal dynamic phenomena which determine the structuring of boiler-turbine control systems, clarifying the essential connections of such phenomena with the physical nature of the process. In this way, it also describes how to capture basic boiler dynamics by simple first-principle models, specifying the limits of their use for control design. Copyright 1997 Elsevier Science Ltd

Keywords: Boilers, control-oriented models, conventional control, power station control, process models, steam plants.

1. INTRODUCTION There is an apparent contradiction between the current practice of power-plant control and the scientific literature proposing advanced control designs for those plants. Engineering practice in power-plant control (Klefenz, 1986; Dukelow, 1986) provides improved solutions by gradual modifications of certain basic control structures, that still appear to constitute the core of the most recent commercial systems. On the other hand, the problem of control-system structuring is often ignored in many papers that consider new control designs based on multivariabie techniques. To cover the gap between "theory" and "practice", it may be very useful to understand why certain control structures (Maffezzoni, 1986) have proved to be so successful in real applications, and to provide conceptual means of identifying possible weak points. Boiler dynamics is just the basis of control-system structuring, since it explains why the fundamental physical processes that develop in power-generation plants determine the relevant interactions among input and output variables, and the nature of the basic dynamic relationships and of the principal nonlinearities (Maffezzoni, 1989; 1990). Investigation of the dynamics of power plants requires detailed models (Maffezzoni, 1992), with representation of plant components; such models are 301

generally used to build plant simulators, where control strategies and controllers' set-up may be assessed (Groppelli, et al., 1992). Large-scale models are generally based on first-principle equations (mass, momentum and energy balances), combined with phenomenological correlations (e.g., heat transfer correlations), and may be considered as knowledge models, i.e. models through which process dynamics can be thoroughly ascertained and understood. Knowledge models are the only reliable way, beside experimentation, to become acquainted with power-plant dynamics, and in particular with the many interactions among process variables and of their actual relevance. However, to justify the basic structure of power-plant control systems, a different kind of model may be very helpful, where only first-cut dynamics is captured, in order to reveal the essential input-output interactions (see e.g. (Astrom and Bell, 1993)). Such models may be called interpretation models; their distinctive feature is that of being based on extremely coarse lumping of mass, momentum and energy balances, whose selection is essentially guided by previous knowledge of the fundamental process dynamics. In light of this, interpretation models base their credibility on having been derived from and compared with knowledge models, and should be considered as a useful tutorial tool to explain fundamental dynamics.

302

C. Maffezzoni
SPRAY DESUPERHEATERS Ist 2 nd SUPERHEATER TSUPERHEATER : ! .--

In this presentation, where the scope is not modeling by itself, but modeling to support control structuring, only simple interpretation models will be developed. It is worth stressing that these simple models are not credible as a basis for the dynamic performance evaluation of control systems, as the dynamics they account for are so essential that any input-output relationship is often reduced to first order. Nevertheless, they account for gross process interactions, to which they give an excellent insight. The paper will first summarize the operating principles from which the fundamental dynamics of boiler-turbine processes can be derived. Then, it will focus on a number of simple models that are suitable for explaining control-system structuring. In the fourth section, typical control schemes are deduced from the previously mentioned simple models. Some concluding comments are made about a number of more-complex dynamic phenomena, especially affecting low-load operation.

Wdr
X~ REHEATER

DRUM ~/~ -(-~

| [

'~H P T L ! J

& i LPT

i I

FR

ED-

STEAM EXTRACTIONS

CIRCULATION LOOP ! FROM L FEED-WATER CYCLE

TO FEED-WATER CYCLE 4

CONDENSER

Fig. 1. Steam-water subsystem To understand the essential input-output relationship in the boiler-turbine process, the conceptual block diagram of Fig. 2 may be useful. Here, the effect of the feed-water cycle on SWS is taken into account by including the feed-water enthalpy among the SWS inputs. There are some drastic simplifications in the scheme of Fig.2. Thermal powers released from the hot gas to the SWS walls is not totally independent of the SWS state (i.e., of wall temperatures); there is almost full independence of Qev (since heat is transferred by radiation from very hot combustion gas), while QSH and QR/-/are more sensitive to wall temperature, because the combustion gas is progressively lowering its temperature. Even more sensitive is QECO, where gas temperature is quite low. However, since the economizer plays a minor role in boiler dynamics, the scheme of Fig. 2 is substantially correct, in the sense that the dominant inputs affecting the thermal powers transferred to the SWS are the fuel and air flows. Due to massive storages of mass and energy, the dynamics of the SWS is much slower than the C&AG dynamics. C&AG dynamics is only relevant when considering specific control and dynamic problems regarding either flame stability (Kwatny and Bauerle, 1986) or the control of the furnace pressure pg (Corno, et al., 1984). Therefore, in most control p~oblems, the C&AG system may be considered together with its feeding systems, as a non-dynamic process part whose crucial role is that of determining power input to the different sections of SWS. In this regard, it is important to identify how C&AG inputs may be used to influence steam generation. It is rather obvious that increasing the total fuel flow into the furnace will simultaneously increase all the heat inputs to the different boiler sections; air flow is varied in strict relation to fuel flow so as to ensure the "optimal" air-to-fuel ratio for combustion.

2. CONTROL OBJECTIVES A fossil-fired power plant is typically needed to supply a certain electric power Pe, while ensuring that the process variables determining process efficiency and plant integrity are kept at their optimal values. The best trade-off between cycle efficiency and plant life results in prescribing certain values to throttle pressure PT and temperature TT and to reheat temperature TR. Moreover, proper operation of the evaporation section requires that the steam separator works in correct conditions, that means with a specified value of the water level YD in the drum. Overall efficiency is substantially affected by the combustion quality and by the waste of energy lost with the flue gas; those requirements combine also with stringent constraints on the environmental impact and, together, they are generally controlled by a proper selection of the air-to-fuel ratio. In coalfired units, finally, the furnace needs to be operated slightly below the atmospheric pressure to minimize the soot dispersion to the environment; in those cases, furnace pressure requires careful control, integrated properly with combustion control. In the light of the complexity of the control objectives and in spite of the multivariable nature of the process, the core of the control system consists almost entirely of independent control loops. To limit the scope of the discussion, reference will be made to drum boiler units and the control aspects relative to the steam-water subsystem (SWS), which has the typical structure of Fig. 1. For those units the most common operating strategy is throttle pressure controlled at a constant value during load variations; however, the modelling approach used to understand fundamental process dynamics can also be used for (controlled) sliding pressure operation (Maffezzoni, 1986).

Boiler-Turbine Dynamics in Power-Plant Control wr I

303

AIIR-GAS

I vl
_ I

WATER
P

~DYNAMI(
otrmER

"l:,

Fig. 2. Input-output structure of the process (Legend: w a , w f = air, fuel mass flow-rates; Ql~co = thermal power to ECOnomizer; QEV = thermal power to EVaporator; QSH = thermal power to SuperHeaters; Qed-t= thermal power to ReHeater). Because of the non-linearity of the heat transfer phenomena, the heat rates QEco, QEV, QSH and Qtut do not vary in proportion to the fuel input: for instance, when the fuel input is increased, the heat rate to the evaporatm: Qgv (that is released in the furnace) increases less than the other heat inputs (Qeco, Qsn, Qva-1)which are released in the furnace backpass. So, while raising the steam generation, steam superheating and reheating would generally increase if proper corrective controls were not applied to rebalance the heat distribution. Temperature control means are then used, either acting on the C&AG system (e.g., the recirculation of combustion gas, influencing the ratio between QEg and the rest of the heat release), or acting on the SWS (e.g., spray desuperheating). From the above discussion it should also be clear that modulation of heat rate to the furnace and variation of the recirculation gas flow simultaneously affect all the process variables, as they influence the heat release to all the different sections of the SWS. This is actually the principal source of mutual interactions in the process. 3. MODELS FOR STRUCTURAL ANALYSIS

The first three assumptions can roughly be considered as low-frequency approximations since, excluding rapid pressure variations, the water subcooling at the downcomers' inlet is quite small. Moreover, because of the very high value of the heat transfer coefficient in the risers (in the order of 100 kW/m 2 K), the metal wall very quickly follows any temperature variation of the fluid. Finally, steam quality is nearly linear at steady state because the heat flux to the furnace wall is evenly distributed. The last assumption is based on the fact that pressure differences along the loop (which are essential for circulation) are of the order of 1% of the absolute fluid pressure in the drum; so the pressure in the evaporator may be identified with the pressure PD in the drum. Thus, the global energy balance and the global mass balance in the evaporator may be written as follows:

dEEz(Po'a)=wwhe-wvhz(Po)+QE dt dM EV(pD,Ct) dt = w w - wv ,

V , (1)

(2)

where w w is the feed-water mass flow-rate (mfr), w v is the steam mfr at the drum outlet, h E is the water enthalpy at the economizer outlet, h vOg) is the steam saturation enthalpy at the pressure p, QEr" is the heatrate to the evaporator, EEV is the total energy stored in the evaporator (fluid and metal of drum, downcomers and risers) and MEt, is the total mass stored in the evaporator. EEV and MEg , beside obvious geometrical parameters like volumes, depend on two process variables: the pressure PD and the void fraction tz in the evaporator, with ct defined as the ratio between the volume occupied by steam and the total volume of the evaporator. To obtain the fundamental pressure dynamics, eqs (1) and (2) are expressed in terms of dpo /dt and d a / dt, then combined to eliminate d a / dt. The final result is:

EV

dpD - QEV - w v ( h v - h E ) + dt -

(3)

3.1 Pressure dynamics


A very simple interpretation model of evaporator dynamics can be derived with the following fundamental assumptions: (1) the fluid in the whole circulation loop (drum, risers and downcomers) is in the state of saturation; the metal walls in the whole circulation loop are at saturation temperature; the steam quality in the risers is, at any time, linearly dependent on the tube abscissa; fluid pressure differences along the circulation loop are ignored while computing mass and energy storages.

- (w w - Wv)(hL - he - [3r)
where the dependence on PD has been omitted for the sake of brevity, h L is the saturated liquid enthalpy, r = h ~ - h z , t = Pv PL - Pv function of PD and ct given by and

CEV

is a

(2) (3) (4)

dr,
CEv = V,ncmp m dp D + + V { ~ r ( l - f l ) dpVdp~+pe d--'~D ~+ jdhV +(I-a)(rfld, p~L +pL dhL I
,. apo

(4)

1l
J

304

C. Maffezzoni ,ON /,OT ~--PN / PT"; SO one may eliminate the pressure ratio PN / P'r, obtaining from (5): wT = f T ( C v ( x ) ) where fT(') P~T-~TTPT f T ( X ) : O~Tp~ (7) of its

with Ts the saturation temperature, Pv and Pc the vapor and liquid saturation and densities, and Vm,, c m , and p , the volume, density and specific heat of the metal walls; V is the total fluid volume in the evaporator. Equation (3) can be given a very useful interpretation in relation to boiler energy storage. At the steady state, due to eq. (2), w w = wv and QL.v = w v ( h v - hl~ ), which means that the input thermal power is spent for steam generation, P.,g: = w v ( h v ( p ~ ) - h E ) being the power necessary to produce the steam mfr w at the pressure Pp. In transient conditions, assuming that the feedwater is controlled so as to maintain the mass balance (i.e., w~ =w~), the possible mismatch between the thermal input QEV and the steam-generation power Psg is turned into energy stored in the evaporator. From eq. (3), it appears that the rate of energy storage is proportional to d p o / d t , so that CEV is naturally interpreted as the evaporator's energy capacitance. Moreover, looking at (4), one may notice that part of the capacitance is due to the storage in the metal walls, and part to the storage in the fluid. For a typical boiler design the metal storage may represent 40+50% of the total boiler capacitance; total drum boiler capacitance may be of the order of 3 MJ/m 3 bar. While hi; can be considered a slowly varying exogenous variable for the evaporator, w v depends on the dnun pressure Pl) and on the total hydraulic resistance opposed by the cascade of superheaters and turbine. First, it is desirable to describe the turbine. Assume, for the sake of simplicity, that the turbine control valves are governed in full arc mode (i.e., with parallel modulation); then the control stage of the turbine (generally of impulse type) can be seen as the cascade of a throttle valve and of a nozzle; this implies: WT = CV(X) 9~TP~ )~v(PN / PT) ) , (5) <6)

is a monotonic

function

argument. Equation (7) tells that the cascade of the turbine and of its control valves behaves like a choked-flow valve with a "special" opening law f T ( x ) . Even when the turbine control valves are commanded in partial-arc mode (i.e., with sequential opening), one arrives at a flow equation of the same type as (7), but with a different form o f the function fT(X) (see (Kalnitsky and Kwatny, 1981), for a more elaborate model of this case). To obtain flow conditions at the evaporator outlet, the flow through superheaters (see Fig. 1) must be described. First-cut modeling of superheaters' hydrodynamics is based on the following remarks: (1) mass storage in the superheaters is very limited because steam has low density; so w v=w r-was ; (8)

(2) head losses in the superheaters develop in turbulent flow and spray mfr Wds is small compared with Wv, so one may write w~. PI~ - Pr = ksn - , PH S (9)

where PSH is a suitable mean density of the superheated steam, and ksH is a constant. Equations (7)-(9) can be combined with eq. (3) to build a simple model of the fundamental pressure dynamics. To this end, a linearized model is derived for small variations about a given steady state, identified as follows: the unit is usually operated at a constant throttle temperature (TT=T~.),sothatthe

wT=KN~NPNZN(p'/pN

where w T is the turbine inlet mfr, C v ( x ) is the flow coefficient of the control valves set (dependent on the valves position x), PT is the steam density at throttle, PN is the nozzle inlet pressure, Zv (fl) is a suitable function of the valve pressure ratio, K N is a nozzle flow constant, ,o N the density at the nozzle inlet, p ' the pressure at the nozzle outlet and Xx a function similar to Z~,- Since the HPT consists of many cascaded stages, it turns out that the pressure ratio ( P ' / P N ) across the control stage will remain nearly constant while varying the flow w T. Moreover, superheated steam behaves very like an ideal gas, and valve throttling is isoenthalpic,

mean superheating temperature TSH at any steady state is nearly equal to its nominal value T~.~/; according to the model o f Section 3.4, the load L (in p.u.) of the plant at any steady state equals the ratio between w T and its nominal value w" " 7

Moreover, superheated steam is assumed to behave like an ideal gas (PsH = PsH / RTsH "~ Pr / RTsu, /gT = P r / R T r , where R is the gas constant) and desuperheating spray mfr in nominal conditions is O assumed to be zero, so that w~' = wT..

Boiler-Turbine Dynamics in Power-Plant Control The model will be expressed in p.u. variables by defining:

305

boiler design; typical values in nominal conditions, are: rev = 200s, c I =1, c 2 =0.1, c 4 = 0.1+0.15. There is a seeming inconsistency in Fig. 3, as the variable Y is considered as an input variable, while its definition implies that it depends, not only on the control variable x, but also on the throttle temperature T T However, it is a common practice to equip the turbine control valve with a wide-band feedback loop of the type shown in Fig. 4. Since valve servo-motors are very fast, and no other lags are present in the loop, at any frequency of interest for the model of Fig. 3 Y=Y. So the turbine admittance actually becomes a control variable, and the loop of Fig. 4 serves for two complementary purposes: it linearizes the nonlinear characteristics of eq. (7), and eliminates the effect of temperature fluctuations on the steam flow to the turbine.

8p = Ap / p~, for any pressure p; 8w = Aw / w~ for any mfr w; 8h=Ah/(hv(PD)-h~)


for any

enthalpy h; 8 T = A T / T for any temperature T expressed in K; 8Qe~= AQ,~ / Q"~ (o denotes nominal conditions). Then eqs (3), (7)-(9) yield the linearized system shown in Fig. 3, where
m o o

xEV = C E v P r / Q e v ,
CI =

he - hE o o h v -h~ dhv Pr dp o h~, - h O ' E


h L - h e - flf

C2 = -

C4 =
/.tp=(l+

OTHER VALVES' POSITION/INPUTS

V
~)p = ~ g T / p ~ ,

w
1~----~

yOL2 / t~p)I ,/z r =2?' o L/Jp, 2-

" ~- "1

I -t ACTUATORS ~

........

P,"

y=(pn-pr)/Pr ,
turbine

Y=fr(x)/
(6)" =

R~-~r
AY/Y ),

is

the

"admittance"

the
I Y

linearization steady state being denoted by the superscript -

d~QE~v E ~h

Fig. 4 Turbine admittance feedback loop From the scheme of Fig. 3 it appears that the pressure dynamics is characterised by a time constant rp, given by:

rev~p

"l'p=

L(]dp(Cl

-C4)-C2~dP

which is about 1.4rev at full load (L=I, ~/, = 1). If the boiler is operated at constant pressurePr , then rev-_--const, and rp is approximately proportional to

Fig. 3 Block diagram of the linearized pressure dynamics Note that yO is usually about 0.05 or less, c 1 is very close to 1, and c 2 is positive for Po > 30 bar and is generally small because he(p) is a fairly fiat thermodynamic function; moreover the effect (very small) of dTsn in eq. (9) has been ignored. The feedwater enthalpy h e undergoes limited and slow variations, so that a* e is a "small disturbance". Finally note that c 4 is usually positive, because feedwater has a significant subcooling, and is structurally small. As the model is expressed in p.u., the above parameters are fairly independent of the

1/L. If the boiler is operated at pure sliding pressure (i.e. ~ p = L), then the variation of rp is mainly due
to rev, which significantly increases at low pressures (e.g., for L= f2p = 0.5, fez may be about 70% larger than at full load). So, in any case, as the unit load increases the pressure dynamics becomes slower. This gross behaviour is well confn-rned by the experimental tests available.

3.2 Drum-level dynamics


Having computed drum pressure by using Fig. 3, eq. (2), which establishes the global mass balance in the evaporator, may be linearized as follows:
type~9 D - o ' a A c t = t~l: w - ~JWv,

(10)

306 with

C. Maffezzoni

OP

= vpl,
w~

l-

dpi
ap

dpv
ap '

Ctr= (1+~)[1-~-rIn[1+~3 ]
with/3 = Pv / (PL - Pv )

(15)

The mfr w r is obtained from the momentum equation

applied to the circulation tubes: Note that, if the power plant is operated at constant throttle pressure, cra is nearly independent of the load, and o-p only slightly dependent on it. To determine the level in the drum, one may write:
a=(V,a, +Vnao)/V , C t r ( P L - P v ) = W2kF(l+ ~ . r X r / ~ ) / P L , (16)

(11)

where k F and 2 r are suitable constants yielding the friction head losses in the downcomers and risers. Equations (15), (16) may be used to eliminate w r and x r from (14), obtaining the following linearized model (written in Laplace transform form): Aetr _ I+sT21[~.2(SQEV_XctSSpD)_~.16PD ] (17) where re, is a normalized capacitance similar to rEv in eq. (10) but relative to the circulation tubes only (typically re, = 0.7 rev), T2 is a small time-constant (of a few seconds) associated with the dynamics of the void fraction within the risers, and 22 and 2 I are positive constants. The difference 6Q.ev - rc, s ~ n has an interesting interpretation: it is the heat rate available for steam generation in the risers, given by the sum of the input thermal power 6QE v and the power released in the case of a decrease in pressure, and corresponding to a reduction in the stored energy. By the L-transformation of eq. (13) and substitution of eq.(17), it f'mally results that

where Vr and VD are the volumes of the risers and of the drum, respectively, while ct~ and a D are the separate void fractions relative to Vr and VD. If it is assumed that, at the considered steady-state the level YD is equal to the drum radius Rz> the result is:
AOt D = -2t~y D ,

(12)

with gYD: = AyD / RD .

Combining eqs (10), (11) and (12), the following equation is obtained:

~Pn = l ( S w w - ~w~) + kp~pn + k~aa~ (13)


TL

-Xct~J~ o

with

"cL -

w~

r[

, kp =

o a 2 VD

--

,k r -

1 5y O = sx L

(SWw-SWv)+ k2
l-sr~ _

I + s T2 5 Q e v +

2 Vo + k I 1-'~sT26PD

,(18)

Equation (13) indicates that the drum level is subject to three different kinds of variations: the first one, of integral type, is due to the imbalance between feedwater mfr and steam mfr; the second one, of proportional type, is due to the dependence of the mean fluid density on the evaporator pressure PD; and the third one comes from possible variations of the void fraction in the risers and might be very quick because any variation Aa r immediately affects d;yD. To determine Aar recall the assumptions at the beginning of Section 3.1, write a couple of equations similar to (1) and (2) but limited to the circulation tubes (i.e. downcomers and risers), and combine the energy and mass equations as follows:
dEct(PD'tr) dt hi ` dMct(pD'c~r) _ dt ,

where
kz

The parameters of model (18) are dimensionless; typical values at nominal pressure are: kp ~.0.7, r L =130s, k z ~ O . 4 , k 1~0.4, T2 ~ 4 s , r c t ~ 1 5 0 s . Since Tz << re,, the time constant T~ ~ 1 5 0 s is always positive and is essentially determined by the "capacitance" re,. The model (18) clearly accounts for the well-known "shrink and swell" effect due to the non-minimum phase zero ( 1 - sT1). Because T2 is very small, any perturbation which gives rise to sudden pressure derivatives causes a sudden variation of the drum level in the opposite direction with respect to the long-term trend. Using the scheme of Fig. 3 and eq. (18), the response of the drum level to a step perturbation of 6 Y can be computed; at nominal load (L=I) and, with the typical values reported in Sections 3.1 and 3.2, the response is plotted in Fig. 5.

(14)

=QEV-XrWr(hv -hL)

where Ect is the total energy (fluid + metal) stored in the circulation tubes, Mct the corresponding fluid mass, x r and w r the steam quality and the mfr at the risers' outlet. Then, from assumption (3) stated at the beginning of Section 3.1, it follows that

Boiler-Turbine Dynamics in Power-Plant Control


Turbine o.o2 ~
o

307

admittance step (bY=O.l) , :


" .............. - ..............

!
! ..............

[ i i i i i i i ~ ~ i .............1 4 - .............. i i i i i

GT(S)

tl@

--

ATOX

q~

-o.o2

.......... ~.............. i .............. ~ .............. i ..............

-o.os ~1

..............~..............i.............. ... .4 ..............i..............

Aw

A Q x t,,[ GQ(S)
Time (s)

Fig. 5 Drum-level response

Fig. 6 Conceptual scheme of temperature dynamics. Characterizing the transfer functions G r (s), G w(s) and G O(s) is relevant to control design. It is known that adequate modeling of superheaters and reheaters requires a distributed-parameter approach; however, it turns out (Keflenz, 1986) that reasonable approximations by lumped parameters may be used for G r, G w and G~). In other words, G w and GO) behave like first-order transfer functions, both with a dominating time constant not far from rEs. The function G r (s) is, in contrast, well approximated as follows:
1

3.3 Reheat and superheat steam-side dynamics.


Superheaters and reheaters are large heat exchangers with steam flowing into the tubes, and gas crossing the tube banks in cross flow. There are some general properties that are worth recalling: (1) the heat transfer coefficient on the gas side is much smaller than that on the steam-side, so that steady-state behavior is nearly independent of steam-side coefficients; (2) the dynamics of these heat exchangers is essentially due to the considerable energy storage in the metal wall, because flue gas has negligible density and steam has much lower capacitance than the corresponding metal wall; (3) mass storage determines much faster dynamics with respect to energy storage, because only steam is involved. Property (3) can easily be checked, bearing in mind that the fundamental time constant of mass storage is r M s = M v / w v, where M~ is the mass of steam within the heat exchanger and "w the rnfr flowing through it, whereas the fundamental time constant of

Gr(s) ~ gT (l+SXE s / N) N ,

(19)

where N is the integer nearest to Sty ,/2wxc p (with

S i and y, the steam-to-waU exchange surface and heat transfer coefficients) a n d / z r is something less than 1. Typically, secondary superheaters are rather short heat exchangers, and have N * 2. Reheaters, which are larger, may have N ~ 3 + 4. Moreover, the process is nonlinear, because rEs is nearly proportional to the inverse of the load, while N is only slightly dependent on Wx, since y, - Wx 8.
So temperature dynamics are affected by multiple lags, which vary with the load; in addition, transducers for steam temperature are generally affected by a small (a few seconds) and a larger (some tens of seconds) time lag due to the thermal inertia of the cylinder where the sensor is placed. Of course, multiple lags are in the loop when AT,x acts as a control variable. This is the case, when desuperheating spray is used (see Fig. 7; refer to the superheating section). Since the attemperator has a very small volume, storages in it are negligible and
W v + W&. = W T

Mvc ~ + MM% , where c v and cpw~ are the specific heats of steam at constant volume d pressure, and M m and c m the mass and the specific heat o f the metal wall. Since M , c , >> M~c~ and cp ~ 1.3c v, it turns out that rEs >> rMs. For a typical superheater rMs is of a few seconds, while res > 20rMs. Mass storage may be ignored when considering temperature dynamics.
energy storage is rEs =

Superheater or reheater outlet temperature Tox (subscript x stands for "generic bank") is essentially influenced by three different variables: the heat rate Qx to the external wall, the steam flow wx and the inlet temperature ~ . Pressure fluctuations within the heat exchanger have a limited influence on T,,x and may be ignored. The situation, for small variations, is described by Fig. 6.

wvho + wd,has = wrh i .


In normal plant operation the steam flow w T in the secondary superheater is imposed (over a wide band) by the load controller, h0 is determined by the upstream superheater and hds is nearly constant.

308

C. Maffezzoni I'm = PHP + PLP


PHe = a r w r ( h r -htR) PL? = aRWR (hR - hc )

FROM

PRIMARYWv 41"
ho "

WT
, '
' ','

TO SECONDARY

(21)

SUPERHEATER

,
% ' "' "

'
"

-~'SUPERHEATER

(22)
(23)

a,

hd~ [ Wds

[hd,,ho,

Fig. 7. Desuperheating spray So the 2nd superheater inlet temperature T i is given by the following variation equation:
ATi

= (ho-=hi) A w r +
Cp WT

_Wv AT

(-ho-hds)_

Awas

WT

Cp WT
(20)

where (refer to Fig. 1) Pnp and PLP are the mechanical power released by HPT and RhT & LPT, respectively, w T is the HPT mfr, h T and htR are the corresponding inlet and discharge enthalpies; w R is the R h T mfr, h R and h c are the corresponding inlet and discharge enthalpies, and a r, a R are suitable constants (< 1) accounting for steam extraction. With the aim of capturing the fundamental process dynamics, one may observe that the enthalpy drops (h r - h t R ) and (h R - h c) remain more or less unchanged as the plant load varies, since turbines are designed to work with constant pressure ratios across the different turbine stages, while the steam flow varies. Thus Equations (21) - (23) can be approximately linearized as:
6Pe = 5Pm = k H p r w T + kRHrWR,

where AT0 = Ah0 /Cp, and the superscript - denotes the steady-state of linearization. In eq. (20) Away is the control variable which directly modulates temperature T/, while Aw r and AT0 are disturbances. Since the influence of Awa, on the pressure is very small, temperature control via desuperheating spray does not significantly influence boiler pressure. In Fig. 6, the heat-rate AQx may also be used to control temperature. This would be very effective because GQ(S) incorporates fewer phase lags than
G r (s). Unfortunately (see Section 2) it is impossible to modulate heat-rate to a single heat exchanger in the boiler without simultaneously influencing all the other heat exchangers. This fact generates interaction among the different process variables, to the extent that it is often necessary to introduce feedforward decoupling actions to achieve acceptable control performance.

(24)

where Flow w T is given by the scheme of Fig. 3. To understand what factors influence WR, refer to Fig. 8. Since HPT has a negligible storage, it is possible to write:
w.r = w , + w.~,.

(25)

When spray is used to control the superheater temperature, then Aw x and AQx are disturbances for the temperature control, due, for instance, to the variation of fuel mfr required by load-pressure control. Fortunately (see Fig. 3), when the heat rate to the evaporator is increased, the steam generation is also increased so that Aw x and AQx grow in nearly the same manner; since Gw(s ) and G~(s) are also similar, then ADx is much smaller than the two individual disturbances, yet (significantly) ADx ~ 0.

The reheater is a large steam heat exchanger; feedwater heaters (FWH) fed by steam extractions are large tube and shell heat exchangers where steam extractions are condensed to heat feed-water. Both components have significant steam mass storage. The FWH represented in Fig. 8 accounts (in an equivalent way) for the overall capacitance of the various FWHs. Ignoring the desuperheatmg spray (normally zero), the relevant mass balances are seen as follows:
dM R dt
- -

= w8 - wR ,

(26)

dMFM dt

= w.,., - w ~ . ,

(27)

where M R is the steam mass in the reheater, and MFH and w c are the steam mass and the condensation mfr in FWHs. Pressure losses in the reheater are small and can be ignored and since VE is normally fully open, it may be assumed that the pressure in the entire reheater is the same as at the RhT inlet. Moreover, reheater temperature has much slower dynamics than mass storage (see Section 3.3), so that it may be considered as constant while evaluating d M R / d t .

3.4 Power generation

When the electric generator is connected to the grid, rotor speed is nearly constant, so that electric power Pe equals mechanical power Pm" Moreover, the turbines have very little storage capacity, so that, ignoring high-frequency effects, turbines may be described by their steady-state equations:

Boiler-Turbine Dynamics in Power-Plant Control

309

_~_

w = mass flow-rate
X

realizing that the electric power output is about 1/3 (kHe ~,0.3+0.4), strictly proportional to the steam flow, and about 2/3 (kHp + ken = 1) is affected by a time lag r R, which cannot be ignored when large quick load variations have to be followed. Finally, observe that the parameters of model (31) are almost independent of the plant load.
3.5 Concluding remarks on process dynamics

Steam L~--it, VV - ~ ~ extractions, ~ - x \ REHEATER


Wse L~~

~'~

su r atin
CONDENSER

FEED-WATERHEATER Fig. 8 Steam reheating

Applying an equation similar to (6) to RhT's first nozzle, one has:


WR = k'RhTP B / ~ ,

To synthesize the analysis of the preceding sections, it is useful to identify the interactions existing in the system. Consider the input control variables, namely the fuel mfr w f , the air mfr wa, the turbine admittance Y, the feed-water mfr Ww, the recirculation gas mfr Wrg, desuperheatmg spray mfr Wds and Wdr, and the output variables to be controlled, namely electric power Pe, throttle pressure PT , drum level YD , throttle temperature T T , reheater outlet temperature TRH and air-to-fuel ratio 2/. From Fig. 3 and eq. (31), it may be seen that Pe and PT are strictly related and are simultaneously affected by heat rate (i.e. wj0 and turbine admittance Y. Flow rates Wds and Wdr very slightly affect Pe and PT because they have only a small mass effect on the main steam flow. Similarly, feed-water flow w w only slightly affects power and pressure (see Fig. 3 and eq. (29)). Air flow is nearly proportional to the fuel flow, so that trimming actions to optimize Aaj. have little influence on the heat rate QEv to the evaporator. Only Wrg changes the heat release partition in the boiler; tlie control band width of such a variable (used to control reheat temperature) is, however, rather narrow just for that reason. Therefore, pressure and power form a (2x2) subsystem, tightly coupled but with limited disturbance from outside. Drum level YD is the only output variable markedly influenced by feed-water flow Ww; YD is also "disturbed" (see eq. (18)) by steam flow Wv, by drum pressure PD and by evaporator heat-rate QEV: so w w weakly influences the load-pressure subsystem, but YD is considerably influenced by the control variables of that subsystem. Superheated steam temperature TT, according to Fig. 6 and eq. (20), is influenced by Wds and is "disturbed" by steam flow and heat rate QSH, which, in turn, follows fuel flow variations. So disturbance of power-pressure subsystem on temperature is relevant, even though there is a natural partial compensation due to the boiler behavior (see ADx in Fig. 6). Reheater spray follows a similar rule to that for superheater spray. Finally, it is worth mentioning that, in coal fired units, fuel is supplied by special pulverizers, which are not influenced by the rest of the plant, but which generally have sluggish dynamics due to the dead time of the grinding process and to the unpredictable effects of variations of coal quality and of

(28)

where TRH is the reheater outlet temperature, R the gas constant and k'~r a suitable constant. Then, taking variations of (25)-(28): 5w R = 1

l+s~R/,w~

/ w~- 5w r _ tlSww) ,

(29)

where r R is a time constant that results from the sum of the storage capacitance of the reheater and of FWHs, multiplied by the flow resistance of RhT, while the last term results from the consideration that condensation mfr variation Aw c is essentially due to feed-water mfr variation Aww. In controlled conditions 8w w strictly follows 8 w r , so that (29) becomes: 1 5w R , ~ - - , S w T (30)

l+s~ R

Equation (29) requires further explanation: the time constant vR takes values of about 10 + 12s, nearly 50% due to the FWH's capacitance (Colombo, et al., 1983); the possibility of varying the principal steam flow Sw n by acting on the feed-water flow mfr3w w has resulted in one of the most recent expedient ways to realize quick power variations even without HPT throttling reserve (Fuetterer, et al., 1992). When the said power control via feed-water is not applied, eq. (30) can be substituted in (24), yielding:

8Pe ~ kHp +

l+sx R

8w T ,

(31)

310

C. Maffezzoni control system of Fig. 9, where K f and /Cf suitable constants. are

component wear (Cao and Rees, 1995). For those plants, fuel flow cannot be considerd as a directly manipulated variable; so the slow response of coal pulverizers must be cascaded with the evaporator dynamics of Fig. 3: this is often the source of severe problems for system stability and control.

Y Ped~

4. MODEL-BASED CONTROL
4.1 Load, pressure and level control

Load Controller

~Y~Admigtance ~

o.o, tTo-rb,Fuel
demand

valve control

PFd

The structure of a power plant control system is conceived,as a combination of feedforward-like actions aimed at balancing inputs with load demand (achieving at the same time the maximum decoupling of control inputs) and of stabilizing/trimming feedback controllers. The simple models developed in Section 3 will now be used to understand the control structure, with reference to the case of units operated at constant throttle pressure. Consider eq. (3), and observe that the last term on the right-hand side is generally small because w w and w v will always be very close to one another and ( h L - h e - f l r ) is a fairly limited enth~lpy drop (with respect to h v - h e ) . Then, to keep the drum pressure constant, the heat input to the furnace should be controlled so as to obtain: Qev = w~(h v - he ) ; (32)

Pressure Feedback Controller

-~
Wvz~

-~

= Measurement

Fig. 9. Basic load-pressure control (dashed block possibly missing; subscript d means demand) Note that the "correcting" signal uf in Fig. 10 is structurally small if the feedforward-like compensation (32) or (33) is properly tuned. To this end, referring to the linearized model of Fig. 3 and eq. (18) and taking into account that c 4 0, feedforward-like compensations for t~Qev and ~nv can be chosen so that 6pr and ~y~ will vanish simultaneously. It is easy to realize that choosing:
8Qzv : pQ(l + sTo)fiw v ,

since it is the throttle pressure PT that must be controlled; in view of eq. (9), a better fuel control would require: with
QEV = wv(hv - h E ) + C E v - ~ t , ~ S H )

(34) (35)

~ww = ( 1 - s T ~ ) ~ v , /.tO = CI -

2L2c2Yo,

TQ = 2LYo rEv - c 4 T ~,

(33)

where the last term on the right hand-side is the necessary overfiring required to compensate for the variations of head losses (9) with the boiler load. However, from Fig. 6 it appears that overfLring (i.e., AQx not balanced by Awx) severely affects temperature; so the trade-off is either to limit load rate (i.e., dw r / d t ) or to allow throttle pressure to vary transiently from its nominal value (Bolis, et al., 1995). Looking now at eq. (31), and bearing in mind that Wr = YPr, if the units are operated at constant throttle pressure, the electric power Pe can easily be controlled, either in an open loop or in a closed loop, via the turbine admittance Y. This is very effective for drum boilers, because CEV is very large, so that pressure PD is a very slowly varying variable. If Pe is controlled in a closed loop over a wide frequency band, then the power loop actually imposes the steam-flow rate w T on the turbine. Then, the law (32) or (33) becomes a quasi-feedforward compensation for the pressure control subsystem. Ignoring the variations of hV - h ~ , one obtains the load-pressure

T ~ = ( k 2 + 2 L T o k l ) r t , makes 8Pr~0 and ~ o negligible. The term s T w in (35) corresponds to the fact that increasing the steam generation requires higher void fraction a r in the risers so that, if the total mass in the evaporator were kept constant (i.e., ~vw = fiwv), the drum level YD would increase with w v . This means that the evaporator inventory should be decreased to keep the drum level constant (Dukelow, 1986). The inventory reduction due to (35) is often realized in practice in the approximate form of a fu'st-order lag:

1 ~w = ~fiw

l+sT~

v .

(36)

A feedback regulator is finally added to stabilize the level dynamics and to ensure steady-state robustness, leading to the popular 3-element scheme of Fig. 10. It is worth noting that the adoption of Tw ~ 0 for the level control not only improves level response to load variation but also reduces the required overfiring sTQ. At full load, typical values are: /~Q ~ 1, Tw ~ 5 0 + 6 0 s and To ~ 10+15s.

Boiler-Turbine Dynamics in Power-Plant Control YDd~o__~Feedback ~ q ~ ~ [ F e e d w a t e r ~_~


To feedpump
co. o,

311

4.3 Remarks

t" IC* r"'rlt


I YI~ t I

t"
I

Ico, o, /

Fig. 10 Level control (subscript d means demand)

4.2 Temperature control

Superheated steam temperature control by water spray (Fig. 7) is based on two basic factors: upslream disturbances (AWTand A T 0 in eq. (20)) are compensated for by means of a tight feedback loop controlling temperature Ti; the final temperature T T is controlled by using the set-point of T i as a control variable, and (19) as the loop process transfer function; the disturbance on the final superheater due to Aw r and AQsa (Fig. 6) may require direct compensation if ADsa is relevant (for instance, when pressure control significantly over-fires).

The power-pressure control is generally implemented in a form that is slightly different from the scheme of Fig. 9; that form is called coordinated control, and is characterised by the use of the pressure error ( P r a - Pr) to limit the load controller demand Yd, when certain prescribed limits are exceeded. The reasoning is as follows: if the boiler is capable of following the turbine loading, then pressure variations should be due only to the head losses of eq. (9); if pressure error definitely overcomes those head losses, it has to be presumed that the loading rate. must be limited. Moreover, to anticipate the coordination of demands to the boiler and the turbine, the feedforward-like action from wv is usually replaced by an equivalent feedforward action from P~I. There remains, however, no loop coupling for small variations. The steam-flow feedback, given by (35) or (36), or even with Tw = 0, has a stabilizing effect on pressure dynamics since, in Fig. 3, generally c4 > 0. At very low loads, however, the economizer may change its behaviour, because flue gas mfr remains quite large while feed-water mfr becomes very small. So the economizer outlet enthalpy hE may arrive close to saturation, and c4 may become smaller than zero. In those conditions the said feedback will have a destabilising effect. This is one of the reasons why the steam flow feedback is usually removed at low loads. Unlike pressure power and drum level, steam temperatures have dynamics of a higher order. This makes the use of advanced controllers of some benefit for implementing R r (Uchida and Nakamura, 1984; Mann, 1992).

This leads to the classical cascade control of Fig. 11. Looking at eq. (20) it may be observed that R,. needs to schedule its gain as 1 / ~ r , while from (19) it results that R r needs to adapt its time constants in proportion to 1 / ~ r . This is usually done in practice. Note that, because control action has a more delayed effect than disturbances, the disturbance compensation and the feedback control cannot limit temperature deviations effectively if the net disturbance ADs, is too heavy. A further problem in temperature control is that the different transfer functions in Fig. 6 are changing with time in a non-predictable way, due to fouling of the banks. This may need periodic retuning of the controller R r.

5. CONCLUDING REMARKS Power-plant control has been developed over several decades, thanks to the contribution of practical engineers working for manufacturers, control suppliers and utility companies. Model-based control was started almost forty years ago with the paper of Chien, et al. (1958). A lot of knowledge has been accumulated since then on boiler dynamics; based on that knowledge, simple interpretation models have been conceived and presented here to understand fundamental boiler dynamics and model-based control concepts. In this paper, the motivations for the principal control-loop structuring have been shown through the given models, limiting the discussion to the basics only. Boiler dynamics exhibit dramatic changes at low loads. Some of the related problems have been explored by means of more elaborate models (see, e.g., (Kwatny and Bauerle, 1986; Maffezzoni and Ferrarini, 1989). In particular, drum-level control must be structurally changed at low loads (Kwatny

fuel ~

flowl

~ steam

I ,ow

compensator t

--tSESOTT

- -

SENSOP~

~Ti

Fig. 11 Desuperheating cascade control.

312

C. Maffezzoni Dolezal, R. and L. Varcop (1970). Process dynamics. Elsevier Publ. Co.. Dukelow, S.G. (1986). The control of boilers. ISA Press. Fuetterer, B., G.K. Lausterer and S.R. Leibbrandt (1992). Improved unit dynamic response using condensate stoppage. Proc. IFA C Symp. Control of Power Plants and Power Systems. Vol. 1, pp. 129-146,. Groppelli, P., M. Maini, G. Pedrini and A. Radice (1992). On plant testing of control systems by a real-time simulator. 2nd Annual ISA/EPRI Joint Control and Instrumentation Conference, Kansas City (USA). Kalnitsky, K.C. and H.G. Kwatny (1981). A frrst principle model for steam turbine control analysis. ASME J. Dynamic Syst., Meas. Control, Vol. 103, pp. 61-68. Klefenz, G. (1986). Automatic control of steam power plants. Bibliographisches Institut, Ztirich. Kwatny, H. G. and J. Bauerle (1986). Simulation analysis of the stability of coal fired furnaces at low load. 2nd IFAC Workshop Modelling and Control of Electric Power Plant, Philadelphia. Kwatny, H.G. and J. Berg (1993). Drum level regulation at all loads: a study of system dynamics and conventional control structures. Preprints XII IFAC World Congress, Sydney. Maffezzoni, C. (1986). Concepts, practice and trends in fossil fired power plant control. Proc. IFAC Symp. on Control of Power Plants and Power Systems, Pergamon Press, pp. 1-9. Maffezzoni, C. (1989). The dynamics of steam generators. Masson Ed. (in Italian). Maffezzoni, C., and L. Ferrarini (1989). Dynamic design of the fast start-up of a Benson boiler. Journal A, Vol. 30, n. 4. Maffezzoni, C. (1990). The control of steam generators. Masson Ed. (in Italian). Maffezzoni, C. (1992). Issues in modeling and simulation of power plants. Proc. IFAC Symp. Control of Power Plants and Power Systems, Vol. 1, pp. 19-27. Mann, J (1992). Temperature control using state feedback in a fossil fired power plant. Proc. IFAC Symp. Control of Power Plants and Power Systems, Vol. 1. Uchida, M. and H. Nakamura (1984). Optimal control of thermal power plants in Kyushu electric power company. Proc. 1st IFAC Workshop Modelling and Control of Electric Power Plants, Como (It).

and Berg, 1993) because the steam flow feedback assumes a destabilizing effect (mostly due to the change of feedwater enthalpy, hE in eq. (3)). Once-through boilers, very popular in Europe, need a completely different representation of the evaporator, and exhibit enhanced interaction between evaporator and superheater control; for these boilers a thorough dynamic analysis is presented in the book of Dolezal and Varcop (1970), while a simple interpretation model was developed by Maffezzoni (1989). In any case, the use of simple models to identify structural properties, in combination with detailed simulators to test the actual performance, is today a credible approach to improving existing control systems, with limited resort to (very expensive) experimental tests. ACKNOWLEDGMENT The work has been supported by CNR-Centro Teoria dei Sistemi and by MURST-Control Engineering Project. REFERENCES ,~,str6m, K.J. and R.D. Bell (1993). A nonlinear model for steam generation process. Preprints of l2th IFAC World Congress, Sidney, Vol. 3, pp. 395-398. Bolis, V., C. Maffezzoni and L. Ferrarini (1995). Synthesis of the overall boiler-turbine control system by single loop auto-tuning technique. Control Engineering Practice, Vol. 3, No. 6, pp. 761-771. Cao, S.G. and N.W. Rees (1995). Fuzzy logic control of vertical spindle mills. Preprints IFAC Symp. Control of Power Plants and Power Systems, Cancun (Mex). Chien, K.L., E.I. Ergin, C. Ling and A. Lee (1958). Dynamic analysis of a boiler. Transactions ASME, Vol. 80, pp. 1809-1819. Colombo, F., A. De Marco, E. Ferrari and G. Magnani (1983). Considerations upon the representation of turbine and boiler in the dynamic response of fossil-fired electrical units. Proc. Cigr~-IFA C Symp. Control Applications for Power System Security, Florence . Corno, D., S. Quatela and A. Zizzo (1984). Dynamic analysis of the air gas path of fossil fuel power plants. Proc. 1st IFAC Workshop Modelling and Control of Electric Power Plants, Como, Italy.

Vous aimerez peut-être aussi