Proceedings of the 19th World Congress

The International Federation of Automatic Control

Cape Town, South Africa. August 24-29, 2014

Plant-wide control of a parabolic trough

power plant with thermal energy storage ⋆
Michael Jost ∗ , Wolfgang Grote ∗∗ , Florian Möllenbruck ∗ ,
Martin Mönnigmann ∗
∗
Automatic Control and Systems Theory, Ruhr-Universität Bochum,
Bochum, Germany
∗∗
MAN Diesel & Turbo SE, Oberhausen, Germany

Abstract: We present a control scheme for a parabolic trough power plant that is equipped
with a molten salt thermal energy storage system. We show that the multivariable control
problem for the solar field can be decoupled by controlling the three way valve that splits the
heat transfer fluid between the storage system and the steam generator. The steam generation
cycle is regulated with inlet pressure control. The subsystems of the plant are modeled based on
first principles, where we focus on the steam generation cycle. A multi-group extraction turbine
with pre- and reheater is considered, which is suitable for solar power plants. We demonstrate
the benefits of the proposed control scheme with a day-to-day simulation.

Keywords: Control of power plants, solar power plants, thermal energy storage, decoupling
control

Technology for solar energy based power generation be- storage system and the steam generator to be an additional
longs to one of two categories: direct usage, most promi- control input. The HTF flow rate is kept constant in this
nently implemented in photovoltaic cells, and indirect setup. This decouples the solar part of the plant into two
usage that is based on concentrating devices, such as subsystems, which can be controlled separately. Surplus
parabolic trough systems (Camacho et al., 2010). The energy collected at daytime can be used to charge the
latter reflect solar irradiation onto an absorber pipe, which storage tanks with a similar control concept based on a
runs in the focal line of a reflector mirror. Solar energy is second three way valve. The power generation block of the
absorbed by a heat transfer fluid (HTF) that flows inside power plant is regulated with inlet pressure control.
the absorber pipe. Finally, the collected thermal energy
is delivered to a steam turbine, which drives a generator. We derive mathematical models for each subsystem of the
solar power plant in Sec. 1. Section 2 details the control
One drawback of these solar power plants is that the main
power source, the sun, cannot be controlled. Obviously, approach. Results are presented in Section 3. The paper
this restricts the operation time of solar power plants. closes with brief conclusions in Sec. 4.
Modern plants are equipped with thermal storage systems
that mitigate fluctuations and that are used to extend the 1. PLANT MODEL
operating hours beyond daylight time. This makes solar
power plants much more attractive, but increases their The solar power plant considered here is equipped with a
complexity from a control point of view. molten salt thermal storage tank (Pacheco et al., 2001).
Figure 1 shows a schematic sketch of the plant, which is
In this paper we deal with the plant-wide control of a solar designed for 50MW electrical power at nominal operation.
power plant equipped with a molten salt thermal storage The steam generator is driven by the HTF flow rate
system. Various controllers have been developed for solar q̇f,3 with inlet temperature Tf,3 . The remainder of the
power plants, such as fuzzy controllers, model predictive plant is a conventional combination of steam turbine and
controllers or internal mode controllers, to name just a few. generator. We use a Rankine cycle to model the generation
We refer to the survey papers (Camacho et al., 2007a,b) of mechanical power. The steam turbine is modeled as
and references therein. a multi-group turbine with reheater and preheater. A
condenser closes the thermodynamic cycle. We summarize
If the solar irradiation is insufficient to run the plant at
the differential equations for each of these subsystems in
its nominal power level, the storage system is used as an
the remainder of this section.
auxiliary energy source. We show how the solar part of the
plant can be controlled in this situation. We consider the
1.1 Notation
three way valve that splits the HTF mass flow between the
⋆ Support by the European Union under the European Regional
We denote by ṁ, q̇, T , ρ, c, A the mass flow rate, volume
Development Fund is gratefully acknowledged. flow rate, temperature, density, specific heat capacity and
cross-sectional area, respectively. Indices m, f, s and w are
used to refer to the pipe metal, the HTF, the molten salt
and the water/steam, respectively. Indices Dis and Chg

978-3-902823-62-5/2014 © IFAC 419

19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

q̇f,1 L2 (µ) q̇f,3 ṁ0 behavior. Assuming that no fluid enters the tank labeled
Dis in Fig. 1 during discharging, the model reads
G ṁs,Dis = −ρs q̇s , Ṫs,Dis = 0,
Dis
q̇f,2 where q̇s is the discharging flow rate. Assuming no fluid
q̇s ṁ3 leaves the storage tank labeled Chg in Fig. 1 during
charging, the model is given by
IN ṁV ṁ5 ṁs,Chg = ρs q̇s ,
q̇s
Ṫs,Chg = (cs Ts ρs − ρs,Chg cs,Chg Ts,Chg ) ,
ṁ4 cs,Chg ms,Chg
Chg where q̇s is the molten salt flow rate entering the tank and
ṁ6 Ts is its temperature.

L1 (λ) HTF to molten salt heat exchanger: We assume a counter-

current heat exchanger is used to transAfer heat between
Fig. 1. Schematic sketch of the solar power plant. HTF the molten salt and the HTF. Applying the first law of
(flow rate q̇f,1 ) is heated in the collector field by solar thermodynamics to this subsystem leads to the differential
irradiation IN . A molten salt storage system can be equations for the temperatures Tf,2,b , Ts,b of the HTF and
the molten salt, respectively, which read
used as second energy source (HTF flow rate q̇f,2 ). The
steam turbine is modeled as a two group extraction Ṫf,2,b = kf,a q̇f,2 Tf,2,a − kf,b q̇f,2 Tf,2,b − k∆T ∆ϑlog (∆TA , ∆TB ), (3)
turbine, where the mass flow rate ṁV is extracted Ṫs,b = ks,a q̇s Ts,a − ks,b q̇s Ts,b + k∆T ∆ϑlog (∆TA , ∆TB ), (4)
after the first turbine group and used to preheat where kf,a , kf,b , ks,a and ks,b are temperature-dependent
feedwater. coefficients that model the thermal properties of the fluids,
refer to the discharging and charging mode of the heat k∆T = kA is a heat transfer coefficient, and indices a and
b refer to inlets and outlets, respectively. The logarithmic
storage, respectively. temperature difference ∆ϑlog (∆TA , ∆TB ) (Incropera et al.,
2007, p. 670 ff.) is defined by
1.2 Physical models of the subsystems ( ∆T − ∆T
A B
, ∆TA ∆TB > 0,
∆ϑlog (∆TA , ∆TB ) = ln(∆TA ) − ln(∆TB ) (5)
Thermal properties of the thermal fluid, the molten salt 0, else,
and water and steam are assumed to be temperature
where ∆TA = Tf,2,a − Ts,b and ∆TB = Tf,2,b − Ts,a .
dependent. See App. A for a brief summary.
Mixers: Assume two fluids with mass flows ṁf,1 , ṁf,2 ,
Collector field: According to Powell and Edgar (2011)
specific heat capacities cf,1 , cf,2 and temperatures Tf,1 , Tf,2 ,
and Camacho et al. (2007a, 1997), the collector field can
be described by the partial differential equations respectively, are mixed. Then the mass flow ṁf = ṁf,1 +
ṁf,2 and temperature
∂Tm
ρm cm Am = η0 DIN − GHl (Tm − Ta ) − LHt (Tm − Tf ), ṁf,1 cf,1 Tf,1 + ṁf,2 cf,2 Tf,2
∂t (1) Tf = .
ρf cf Af
∂Tf
+ ρf cf q̇f
∂Tf
= LHt (Tm − Tf ),
ṁf,1 cf,1 + ṁf,2 cf,2
∂t ∂x result at the outlet of the mixer.
where IN denotes the solar irradiation, D is the mirror
width, G is the inner diameter of the pipe, L is the Valves: Consider the three way valve L1 (λ) with inlet mass
outer diameter of the pipe, η0 is the efficiency, Ta is the flow and temperature ṁf and Tf , respectively, which splits
ambient temperature and Hl (Tm ) and Ht (q̇f , Tf ) are the the HTF flow rate into ṁf,1 and ṁf,2 . Let λ ∈ [0, ..., 1] be
heat exchange coefficients. Note that these coefficients the valve parameter. Then the outlet mass flows are given
are not constant but depend on the current pipe and
fluid temperatures Tm , Tf , respectively. We adopt these by
coefficients from (Camacho et al., 1997, p. 29 ff.). We ṁf,1 = λṁf , ṁf,2 = (1 − λ)ṁf .
(t,x)
discretize (1) using ∂Tf∂x ≈ Tf (x,t)−T∆x
f (x−∆x,t)
for small We assume the thermal properties of the fluids do not
∆x. Setting Tf (x, t) = Tf,i and Tf (x − ∆x, t) = Tf,i−1 , this change across the valve, which yields Tf,1 = Tf , Tf,2 = Tf .
yields
ρm cm Am Ṫm,i = η0 DIN − GHl (Tm,i − Ta ) − LHt (Tm,i − Tf,i ),
Steam generator: The steam generator model consists of
Tf,i − Tf,i−1 (2) three subsystems: A preheater is used to heat the liquid
ρf cf Af Ṫf,i + ρf cf q̇f = LHt (Tm,i − Tf,i ), water to the saturated liquid line, then an evaporator
∆x
is used to model the change of the enthalpy in the wet
where we assume that the same solar irradiation IN applies steam region and finally, a superheater is used to model
to all mirrors. The collector field considered here contains the enthalpy change in the generated steam.
312 rows of absorber pipes of 250m length each. We
discretize each row into 10 elements (i = 1, . . . , 10 in (2)) Both the preheater and superheater are modeled as coun-
with 25m length each. terflow heat exchangers. We only summarize the results
for the preheater. The energy balance of this subsystem
leads to the differential equation
Molten salt storage tanks: Charging and discharging are
described with separate models. Both models are based ṁf,7 kA
Ṫf,7 = (Tf,6 − Tf,7 ) − ∆ϑlog (Tf,7 − Tw,6 , Tf,6 − Tw,7 ),
on energy and mass balances, where we assume adiabatic mf,7 mf,7 cf

420
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

ṁf,8
V7 Steam ṁst V8
ṁ1 uV
ṁ7 Water ṁ0
ṁ8
p1 , V1
Fig. 2. Model of the steam generation process considered
ṁ2
in this paper. The feedwater is vaporized in the evap-
orator (left), and the generated steam is overheated HP LP G
in the superheater (right). Since no phase transition
takes place in the preheater, this part is not shown
ṁ2 ṁ3 ṁ3
here.
where k is the heat transfer coefficient and ∆ϑlog (Tf,7 − p2 , V2
Tw,6 , Tf,6 − Tw,7 ) is the logarithmic temperature difference
(5). A similar differential equation for the HTF tempera- ṁV
ture in the evaporator can be derived. It reads
2ṁf,8 2kA Fig. 3. Simplified schematic representation of the turbine:
Ṫf,8 = (Tf,7 − Tf,8 ) − ∆ϑlog (Tf,8 − Tw,7 , Tf,7 − Tw,8 ),
mf,8 mf,8 cf After passing the high pressure group a fraction of the
where we have assumed that only 50% of the coils are steam (flow rate ṁV ) is used to reheat the feedwater.
immersed in feedwater. The remainder (flow rate ṁ3 ) is reheated and fed into
the low pressure group.
We describe the pressure and steam generation in the
steam generator next; see Fig. 2 for its layout. The pressure is extracted and used to preheat the feedwater. The
generation in both subsystems is modeled with a finite remaining mass flow rate ṁ3 is reheated using a reheater
volume approximation. This yields and fed into the low pressure group. The mass flow entering
κRT7 z the high pressure group can be controlled with a valve with
ṗ7 = (ṁst − ṁ8 ), characteristics
V7 s
κRT8 z p0 T0,0
ṗ8 = (ṁ8 − ṁ0 ), ṁ1 = ṁ0 (3u2V − 2u3V ),
V8 T0 p0,0
where κ is the isentropic exponent, z the compressibility where T0,0 and p0,0 are the operation point temperature
factor, R the specific gas constant and Vi and pi are the and pressure, respectively (Grote, 2009, p. 78 ff.). Note
corresponding volumina and pressures, respectively, and that the normalized valve position is bounded by 0 ≤
Ti are the temperatures (Grote, 2009, p. 58). uV ≤ 1. The mass flow through the high pressure turbine
A throttle valve is used to control the mass flow ṁ8 from group can be found using Stodola’s law of the ellipse, which
the steam generator to the boiler. The throttle is modeled yields
as a fluidic resistance with a turbulent flow. According to
s
p21 − p22
the Moody diagram (Incropera et al., 2007, p. 491) this ṁ2 = KST , (6)
yields T1
 4
p7 − p8 7 where KST is a constant. KST is only depends on the nom-
ṁ8 = . inal turbine design parameters p1,0 , p2,0 and T1,0 (Grote,
R 2009, p. 45 ff.). It is given by
The enthalpy change of the feedwater and the steam s
in the steam generator are calculated with the IAPWS- T1,0
KST = ṁ0 . (7)
IF97 formulas (Wagner and Kretzschmar, 1997). The total p21,0 − p22,0
change of enthalpy is given by
Since the model of the low pressure turbine group is
∆h = ∆hP + ∆hE + ∆hS , identical to (6) and (7), the equations are not repeated
where ∆hP = h′ (p6 ) − h(p6 , T6 ) is the enthalpy change here. Finally, the dynamics of the pressure after the
in the preheater, ∆hE = h′′ (p6 ) − h′ (p6 ) is the enthalpy high pressure group are modeled with a finite volume
change in the evaporator and ∆hS = c∆T is the enthalpy approximation (Grote, 2009, p. 58 ff.), which yields the
change in the superheater. Finally, the vaporized fraction differential equation
of the feedwater ṁst is given by κRT2 z
kA ṗ2 = (ṁ2 − ṁV − ṁ3 ).
ṁst = ∆ϑlog (Tf,7 − Tw,6 , Tf,6 − Tw,7 ), V2
hE The pressure dynamics after the low pressure group are
where hE is the enthalpy of the vaporized water (Sonntag included in the condenser. For simplicity, the dynamics of
and Van Wylen, 1971, p. 140 ff.). the pipes are neglected.

Steam turbine: The steam turbine is modeled as a two Generator: We model the electrical generator as a syn-
group extraction turbine with pre- and reheater. A sketch chronous machine, which is coupled to an electric grid.
of the turbine is given in Fig. 3. After the steam has The resulting second order nonlinear differential equation
passed the high pressure group, the mass flow rate ṁV reads

421
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

1
(MI − MT sin(ϕ) − de ϕ̇),
ϕ̈ = q̇f,1 = λq̇f,3 Tf,1
θ
where ϕ is the rotor angle, de is a damping constant, MT
is the tilting torque of the machine and θ is the inertia
torque. The power fed into the electrical grid is given by q̇f,2 = (1 − λ)q̇f,3 Tf,2,b
L1 (λ) L2 (µ)
///////
Pe = MT sin(ϕ)2πf0 ,
where f0 is the nominal grid frequency (Chapman, 2012).
P1 q̇f,3 A B
q̇s
Preheater and reheater: The preheater and reheater are
modeled as counter flow heat exchangers. Thermal prop- q̇f,3 , Tf,3
erties of water and steam at the output port are calculated (a) Controller layout during night and sunset/sunrise
with the IAPWS-IF97 tables (Wagner and Kretzschmar, q̇f,1 Tf,1
1997). The preheater model has to take the phase transi-
tion of the extracted steam ṁV into account. Since they are
analogous to those of the steam generator, the equations
P2
are not repeated here. q̇f,2 = (1 − µ)q̇f,1
L1 (λ)
/////// L2 (µ)
q̇f,3 = µq̇f,1
Condenser: Finally, after the steam passes through the
Ts,b
low pressure group of the turbine, it is condensed in a P1 A B
condenser, which closes the thermodynamic cycle. The q̇s
exhaust steam pressure dynamics are modeled with a finite q̇f,3 , Tf,3 = Tf,1
volume again, which yields (b) Controller layout during daytime
κRT4 z
ṗ4 = (ṁ3 + ṁ5 − ṁ4 ).
VC
Note that this volume is assumed to include the pressure Fig. 4. Control of the collector field (collector field out-
dynamics after the low pressure turbine group. The ex- let temperature control loop in red; heat exchanger
haust steam temperature can be calculated with Magnus’s control loop in blue; known disturbance, which is
formula compensated in the controller, in green). (a) Stored
energy is used to drive the plant at night and during
p


4
c2 ln c0
T4 = p4

+ 273.15, sunset/sunrise. If both Tf,1 and Tf,2,b are at steady
c1 − ln c0 state, the temperature of the HTF at the steam gen-
where c0 , c1 and c2 are specific constants (Alduchov and erator is constant, too. (b) Collected heat is partially
Eskridge, 1996). The dynamics of the cooling water simply used to run the plant and partially stored for use at
follow from the first law of thermodynamics and read night/sunset/sunrise.
2ṁw 2kA
Ṫw,2 = (Tw,1 − Tw,2 ) − ∆ϑlog (Tw,2 − Tc,1 , Tw,1 − Tc,2 ). 2.1 Collector field
mw mw cw
We assume that the cooling water level is controlled, and
50% of the coils are immersed in water. Operation at night/sunrise/sunset: Since the energy col-
lected by the field is insufficient to run the power plant at
Feedwater pump: The feedwater pump is modeled as a its nominal power output, the molten salt thermal storage
centrifugal pump. According to Gülich (2010), system has to be used as an additional energy source. We
use the control layout shown in Fig. 4(a). Essentially, the
∆p = d0 + d1 q̇ + d2 q̇ 2 , pump P1 is used to regulate q̇f,3 to the constant value
holds where d0 , d1 and d2 are constants and ∆p is the q̇f,3,nom , the temperature Tf,2,b is regulated to a nominal
pressure difference between the inlet and outlet. The latter value with q̇s , and constant Tf,3 is achieved by controlling
equation can be extended to include the pump speed, the fraction of the HTF that passes the collector field
which yields with the three way valve L1 (λ). We explain below that
 2
this layout is favorable, because it decouples the control
 
n n
∆p = d0 + d1 q̇ + d2 q̇ 2 , of q̇f,3 , Tf,1 and Tf,2,b . We omit the description of the flow
n0 n0
rate controller for P1 and q̇f,3 for brevity and discuss the
where n0 denotes the nominal rotary speed, and n is the remaining two controllers.
pump speed (Leonow and Mönnigmann, 2013).
The system of ODEs (2), which describes the collector
2. CONTROLLER DESIGN field and results from discretizing the partial differential
equation (1), is nonlinear. Linearization results in a high
Nominal operation corresponds to a HTF flow rate order transfer function, which can be reduced to the second
3
q̇f,3,nom = 1.25 ms at a temperature Tf,3,nom = 674K. Flow order transfer function
rate and the temperature must be constant during the Tf,1 −1.26s − 80.63
Gf = = 2 · 10−2 .
entire day to maintain the nominal power output. Differ- q̇f,1 s + 34.4s + 0.42
ent controllers are required at night and during sunrise Since q̇f,1 = λq̇f,3,nom , where λ ∈ [0, ..., 1],
and sunset on the one hand, and during the day on the
other hand to achieve nominal operation. Both setups are Tf,1 = Gf q̇f,3,nom λ (8)
described in Sec. 2.1. We present the controller design for where we note again that q̇f,3,nom is constant under the
the steam generation cycle in Sec. 2.2. assumptions stated above. We use PI control

422
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

Z t
u(t) = KP e(t) + KI e(τ )dτ, (9) 1000 3
0
where u(t) = λ(t) is the controller output and e(t) = 750
1.5
Tf,3,nom − Tf,3 is the control error. The parameters KP and

q̇f,1 , q̇f,2 , q̇f,3

KI are determined such that the closed-loop shows fastest

I˜N , IN
500
response without overshoot using the root locus method. 0
The solar irradiation and the inlet temperature at the 250
collector field are assumed to act as unknown disturbances
on the system. A linear description of the molten heat to 0
6 9.5 13 16.5 20
-1.5
6 9.5 13 16.5 20
t t
HTF exchanger can be obtained from linearizing (3). It
(a) (b)
reads
Tf,2,b = Gf,f (1 − λ)q̇f,3,nom + Gf,s q̇s , (10) 675 1

where q̇s is the molten salt flow rate and (1 − λ)q̇f,3,nom is

the flow from L1 (λ) (see Fig. 4(a)). The transfer functions
Gf,f and Gf,s are given by 674 0

q̇s
Tf,3
−3.32s − 0.43 0.42
Gf,f = 2 and Gf,s = 2 · 10−2 ,
s + 13.47s + 0.48 s + 13.47s + 0.48
respectively. Note that (1−λ)q̇f,3,nom acts as a disturbance
673 -1
on this system. Since λ is given and q̇f,3,nom is assumed 6 9.5 13
t
16.5 20 6 9.5 13
t
16.5 20
to be constant, this disturbance is known and can be (c) (d)
compensated by setting
Fig. 5. Day-to-day collector data: (a) Solar irradiation I˜N
Gf,f
q̇s = − (1 − λ)q̇f,3,nom + q̂˙s , (blue) and the irradiation with random walk distur-
Gf,s bance I˜N (red) between 6AM and 8PM. Grey back-
and considering q̂˙s to be the input. Unfortunately, the ground indicates sunrise and sunset times, when the
compensator transfer function Gf,s
Gf,f
is not realizable. This storage system is used as an additional energy source.
White background indicates storage charging. (b)
problem can be overcome by setting HTF flow rate through the collector field q̇f,1 (blue),
Gf,f 1 steam generator q̇f,3 (green) and the salt to HTF heat
q̇s ≈ − (1 − λ)q̇f,3,nom + q̂˙s , (11) exchanger q̇f,2 (red). Opposite signs of q̇f,2 indicate
Gf,s Tr s + 1
opposite flow directions; negative values corresponds
which yields to charging, positive values to storage discharging. (c)
 
1 HTF temperature at the steam generator Tf,3 is con-
Tf,2,b = Gf,f 1 − (1 − λ)q̇f,3,nom + Gf,s q̇s . stant over the day-to-day cycle. (d) Molten salt flow
Tr s + 1
1
rate through the molten salt to HTF heat exchanger
The term (1 − Tr s+1 ) describes a high-pass filter, which q̇s . Signs indicate flow direction.
damps signals with frequencies ω < T1r . Assuming that
the solar irradiation and thus the HTF flow rate changes
1
slowly, 1− Tr s+1 ≈ 0 holds for Tr sufficiently small and the Consider the linear model of the collector field again. As-
suming that the inlet temperature and the solar irradiation
transfer characteristics simplifies to Tf,2,b = Gf,s q̂˙s . We use are unknown disturbances,
a PI controller, which has been tuned with the root-locus
method, to regulate Tf,2 to Tf,2,nom = 674K. Tf,1 = Gf q̇f,1 , (13)
Note that choosing the valve as the controller input and where now the HTF flow rate q̇f,1 is the controller input.
KI
using the disturbance compensation (11) decouples the By following the same steps as for (8) we find K̃I = q̇f,3,nom
multi-input multi-output system defined by the differential KP
and K̃P = q̇f,3,nom , where KI and KP are the parameters
equations (8) and (10), which describe the solar field and
the heat exchanger, respectively. Decoupling results in the of the PI controller for (8).
two single-input single-output systems The fraction of HTF that is not delivered to the steam
Tf,1 = Gf q̇f,3,nom λ, generator is used to charge the molten salt storage. Lin-
(12) earizing (4) yields Ts,b = Gs,f (1 − µ)q̇f,1 + Gs,s q̇s ,where (1 −
Tf,2,b = Gf,s q̂˙s . µ)q̇f,1 is a known disturbance, and the transfer functions
read
Daytime operation: During daytime, more energy is col- −13.62 3.65s + 23.48
lected in the solar field than necessary for nominal opera- Gs,f = · 10−2 and Gs,s = .
s2 + 6.46s + 0.04 s2 + 6.46s + 0.04
tion. The excess energy can be used to charge the molten
salt heat storage. We use the three way valve L2 (µ) to split This disturbance can be compensated again using the
Gs,f
the HTF between the steam generator and the molten salt compensator q̇s = − Gs,s (1 − µ)q̇f,1 + q̂˙s with the artificial
heat exchanger (see Fig. 4(b)). Since the flow rate to the input q̂˙s . In summary, this yields Tf,1 = Gf q̇f,1 , Ts,b =
steam generator, q̇f,3 , is assumed to be fixed, the parameter Gs,s q̂˙s . In contrast to (12), the salt output temperature is
q̇
µ can be calculated from µ = f,3,nom q̇f,1 . chosen as the second output.

423
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

2.2 Steam generation

Since the turbine is modeled as a multi-group turbine with 109 52

pre- and reheater, and the thermal properties of water
50
and steam are not assumed to be constant, this part of
the power plant is a nonlinear multi-input multi-output

Pe
107 48

p1
system. For simplicity, we assume that (a) the ratio of the
46
steam generator HTF flow rate to the reheater HTF flow
105
rate is fixed, (b) the temperature of the cooling water is 44
constant, (c) the pipes are ideal, and (d) a constant mass 6 9.5 13 16.5 20 6 9.5 13 16.5 20
t t
flow ṁV is withdrawn to preheat the feedwater.
(a) (b)
We apply inlet pressure control to control this part of the
plant. This type of controller stabilizes the pressure p1 1
5
before the high pressure group of the turbine using the
input valve uV (see Fig. 3). The mass flow through the
high pressure group and thus the thermal power output

ṁD
uV
0.8 2.5
varies in this case. The linear input to output dynamics
from input uV to output p1 can be found by linearizing the
turbine, the pre- and reheater, and the steam generator at
their nominal operation points and can be written as 0
0.6
6 9.5 13 16.5 20 6 9.5 13 16.5 20
p1 = G̃SG uV . (14) t t
(c) (d)
Computing a reduced order approximation of (14) results
in the transfer function GSG , which has four stable poles Fig. 6. Results of the steam generation part for the
and two complex-conjugate zeros with negative real part simulated day-to-day cycle: (a) Inlet pressure to the
1.44s2 −9.31s−16.02 7 high pressure group of the turbine is stabilized by the
and reads GSG = s4 +25.56s 3 +126.9s+6.59 · 10 . A standard
controller. (b) Generated power of the solar plant.
PI controller has been used to compensate the slowest pole If the extracted mass flow ṁD increases, the mass
of GSG , and the controller gain was chosen such that the flow through the turbine decreases, consequently the
resulting closed loop shows aperiodic behavior with the generated power decreases. (c) The inlet valve at the
root locus method. turbine is used as the control input: if the mass flow
through the turbine decreases, the pressure at the
3. SIMULATION AND RESULTS turbine decreases. The controller closes the valve to
regulate the pressure back to its nominal level. (d)
We simulate the power plant for an entire day to assess Extracted mass flow ṁD , see (16).
the performance of the proposed controllers. The nonlin-
ear subsystems described in Sec. 1 are implemented in only uniaxial, which leads to an error that can be modeled
C, and an implementation of the IAPWS-IF97 formulas by
in C are used. The simulations are carried out using
Matlab/Simulink. Before discussing simulation results, we I˜N = I cos(Θ).
summarize the calculation of the solar irradiation. The
The inclination angle Θ depends on the solar altitude angle
solar irradiation consists of two contributions, the direct
and the location of the plant. It can be calculated from
and the diffuse irradiation. Since the power output of solar
simple geometric considerations (Quaschning, 2011, p. 63
plants mainly depends on the direct irradiation, we neglect
ff.).
the diffuse component.There exist various methods for the
estimation of the solar irradiation at ground level. We We add a random walk to I˜N to model measurement
choose the European Solar Radiation Atlas (ESRA) clear- errors. More precisely, the solar irradiation used for simu-
sky model. According to Rigollier et al. (2000) the direct lations in this paper is modeled by
solar irradiation is given by
IN = I˜N + ξ(t), (15)
I = I0 ǫ0 sin(ψ)e−0.8662TL mδ
Pk
where I0 = 1367 m W
2 is the solar constant, ǫ0 is a correction
where ξ(t) = i=1 τi (k T̃ ), with a uniformly distributed
term, ψ is the solar altitude angle, TL is the Linke random variable τi (k T̃ ) ∈ [−5, 5], where T̃ = 25s is the
turbidity factor, m is the relative optical air mass and sample time. The interval limits and the sample time have
δ is the integral Rayleigh optical thickness. Both the been chosen such that the effect of the disturbance is
relative optical air mass and the integral Rayleigh optical reasonable.
thickness depend on the altitude above sea level. We refer
Steam generators or steam turbines are often tapped
to Rigollier et al. (2000) and references therein for the
calculation of TL , m and δ. to draw steam for other purposes than electrical energy
generation, for example, for simple heating purposes. Since
The calculated solar irradiation I holds for a surface nor- disturbances of this type are common, we assume the
mal to the inclination angle. In most plants, the collectors steam generator is tapped and steam is extracted at a
track the sun over the day to maximize the collected en- time-varying mass flow rate ṁD (t). Specifically, we assume
ergy. In the plant considered here, however, the tracking is ṁD (t) to be the periodic function

424
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

controller input. Finally, further research should address

  
5 Kg s−1 t T
0 ≤ t− T < set point tracking for the inlet temperature and HTF flow
ṁD = T 2 , (16)

0 Kg s−1 else rate at the steam generator for use in start-up and shut-
down procedures of the power plant and the power cycle.
with period T = 9000s.
Simulation results for the controlled collector field are Appendix A. THERMAL PROPERTIES OF FLUIDS
shown in Fig. 5. Figure 5(a) shows the calculated solar
irradiation as well as the disturbed signal (15). The solar The density and heat capacity of a typical thermal oil can be modeled
irradiation increases, and reaches a maximum around as
noon. As the irradiation increases, more and more HTF is ρf = 1071.0 − 0.6306(Tf − 273.15) − 0.0007646(Tf − 273.15)2 ,
routed through the collector field by the controller. During cf = 1510.8 + 2.254(Tf − 273.15) + 0.00062631(Tf − 273.15)2 ,
this period, the molten salt is used as a secondary energy respectively. For the density and heat capacity of the molten salt we
source (Figs. 5(a,b)). When the HTF flow rate reaches assume
q̇f,3,nom , more heat is collected in the solar field than
ρs = 2265.38 − 0.64Ts ,
necessary to run the plant. The HTF flow rate through the
collector field increases further, and the storage system is cs = 1400.48 + 0.17Ts ,
charged with the additional heat (sign of the molten salt respectively. Thermal properties of water and steam are implemented
flow rate indicates charging). When the solar irradiation based on the International Association for the Properties of Water
drops below the nominal level q̇f,3,nom , the storage is used and Steam (IAPWS-IF97) (Wagner and Kretzschmar, 1997).
as an additional energy source again to run the plant.
Obviously, the temperature Tf,3 and the flow rate through REFERENCES
the steam generator q̇f,3 are constant over the entire day- Alduchov, O.A. and Eskridge, R.E. (1996). Improved Magnus form
to-day cycle (Figs. 5(b,c)), which shows that the presented approximation of saturation vapor pressure. Journal of Applied
control strategy is able to stabilize the collector field Meteorology and Climatology, 35, 601–609.
despite the imposed disturbances. Fig. 5(d) shows the Camacho, E.F., Berenguel, M., Valenzuela, L., and Rubio, F.R.
molten salt flow rate q̇s (1997). Advanced Control of Solar Plants. Springer.
Camacho, E.F., Rubio, F.R., Alvarado, I., and Limon, D. (2010).
Simulation results for the steam part are shown in Fig. 6. Control of solar power systems: a survey. In Proceedings of the 9th
The inlet pressure to the high pressure group is shown International Symposium on Dynamics and Control of Process
in Fig. 6(a). Steam extraction according to (16) leads Systems, 817 – 822.
to a significant disturbance, but the controller is able Camacho, E.F., Rubio, F.R., Berenguel, M., and Valenzuela, L.
(2007a). A survey on control schemes for distributed solar collector
to stabilize the pressure at 107bar. Not surprisingly, the
fields Part I: Modeling and basic control approaches. Solar Energy,
generated power decreases if the mass flow through the 81, 1240–1251.
turbine decreases (see Fig. 6(b)). Figure 6(c) shows the Camacho, E.F., Rubio, F.R., Berenguel, M., and Valenzuela, L.
normalized valve position. Obviously, to increase the pres- (2007b). A survey on control schemes for distributed solar
sure in front of the turbine group after ṁD increases, the collector fields Part II: Advanced Control Approaches. Solar
valve closes. Finally, the extracted mass flow rate ṁD is Energy, 81, 1252–1272.
shown in Fig. 6(d). Chapman, S.J. (2012). Electric Machinery Fundamentals. McGraw
Hill.
Grote, W. (2009). A contribution to model based control of extraction
4. SUMMARY AND OUTLOOK steam turbines (In German). Ph.D. thesis, Ruhr-Universität
Bochum.
We presented a plant-wide control strategy for a solar Gülich, J.F. (2010). Centrifugal Pumps. Springer.
power plant with molten salt storage system. The strategy, Incropera, F.P., DeWitt, D.P., Bergman, T.L., and Lavine, A.S.
which uses the three way valves in the thermal oil cycle (2007). Fundamentals of Heat and Mass Transfer. Wiley & Sons,
as additional control inputs, has been shown to stabilize 6th edition.
the inlet temperature and HTF flow rate at the steam Leonow, S. and Mönnigmann, M. (2013). Soft sensor based dynamic
flow rate estimation in low speed radial pumps. In Proceedings of
generator during a day-to-day cycle with disturbances. In-
the European Control Conference 2013, 778–783.
let pressure control was implemented to control the power Pacheco, J.E., Showalter, S.K., and Kolb, W.J. (2001). Development
generation block of the power plant. Our simulations show of a molten-salt thermocline thermal storage system for parabolic
that the controller is able to regulate the inlet pressure to trough plants. In Proceedings of the Solar Forum 2001.
the turbine in the presence of disturbances. Powell, K.M. and Edgar, T.F. (2011). Control of a large-scale
solar thermal energy storage system. In Proceedings of the 2011
Several extensions of the presented work can be outlined American Control Conference, 1536–1541.
as follows. We used linear controllers that decouple the Quaschning, V. (2011). Regenerative Energiesysteme: Technologie –
nonlinear multi-input multi-output system in Sec. 2.1. Berechnung – Simulation. Carl Hanser Verlag, 7th edition.
The linear models used to derive the controllers are only Rigollier, C., Bauer, O., and Wald, L. (2000). On the clear sky model
valid at the steady state of linearization. The controller of the ESRA – European Solar Radiation Atlas – with respect to
performance may decrease if the plant is not operated un- the heliosat method. Solar Energy, 68, 33–48.
der nominal conditions and the disturbance compensation Sonntag, R.E. and Van Wylen, G.J. (1971). Introduction to Ther-
modynamics: Classical and Statistical. John Wiley & Sons, Inc.
may not be able to decouple the multi-input multi-output
Wagner, W. and Kretzschmar, H.J. (1997). International Steam
system. Moreover, some assumptions applied to the steam Tables: Properties of Water and Steam based on the Industrial
generation cycle can be dropped. For example, we assumed Formulation IAPWS-IF97. Springer, 2nd edition.
a fixed ratio of the steam generator and reheater HTF flow
rates. This ratio can be considered to be an additional

425